{"name":"ModelFit Local Image and Video Model Dataset","description":"Exact downloadable files, component bundles, software routes and published Mac measurements for local image and video models. Unknown runtime costs stay null; a download size is not a measured memory peak.","source":"https://modelfit.io/image/","license":"CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)","attribution":"ModelFit: https://modelfit.io/. Arena quality rows carry their own required attribution.","updated":"2026-10-24","counts":{"models":13,"image":8,"video":5},"routes":[{"id":"argus-mlx","name":"Argus-AiTeam/minimax-h3-mac (MLX streaming)","memoryPolicy":"ssd-streaming","formats":["safetensors","mlx"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"sources":[{"label":"Argus MLX route README","url":"https://raw.githubusercontent.com/Argus-AiTeam/minimax-h3-mac/dffe3d664f2afd9439bd32157e926aa5494175a0/README_EN.md","checkedAt":"2026-10-24","evidence":"Total model size primarily affects disk space and I/O volume rather than peak unified-memory residency."}]},{"id":"comfyui","name":"ComfyUI","memoryPolicy":"resident-sum","formats":["safetensors","gguf"],"platforms":["mac","nvidia"],"support":"official","activationBytes":null,"overheadBytes":null,"note":"On Apple Silicon ComfyUI runs on PyTorch MPS, not MLX. GGUF files need the separate city96/ComfyUI-GGUF custom node. Resident size is the sum of the weights, an upper bound: ComfyUI can unload models between stages and publishes no per-model Mac peak.","sources":[{"label":"ComfyUI README","url":"https://github.com/Comfy-Org/ComfyUI/blob/7cebca52227c63b4180ceed448cbd7ad22903b6d/README.md","checkedAt":"2026-10-24","evidence":"Supports all operating systems and GPU types (NVIDIA, AMD, Intel, Apple Silicon, Ascend)."},{"label":"ComfyUI system requirements","url":"https://github.com/Comfy-Org/docs/blob/f6fb21b37bc7a5d5123b27fd2e93e9e98dedf92e/installation/system_requirements.mdx","checkedAt":"2026-10-24","evidence":"ComfyUI does not use MLX directly."},{"label":"ComfyUI CLI arguments","url":"https://github.com/Comfy-Org/ComfyUI/blob/7cebca52227c63b4180ceed448cbd7ad22903b6d/comfy/cli_args.py","checkedAt":"2026-10-24","evidence":"By default models will be unloaded to CPU memory after being used."},{"label":"ComfyUI-GGUF README","url":"https://github.com/city96/ComfyUI-GGUF/blob/6ea2651e7df66d7585f6ffee804b20e92fb38b8a/README.md","checkedAt":"2026-10-24","evidence":"GGUF Quantization support for native ComfyUI models"}]},{"id":"comfyui-krea2-mlx","name":"ComfyUI-Krea2-MLX","version":"v0.4.0","memoryPolicy":"resident-sum","formats":["mlx"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"note":"Unofficial ComfyUI node pack that runs Krea 2 Turbo in-process on MLX. It keeps one model resident and does not plug into the native KSampler.","sources":[{"label":"ComfyUI-Krea2-MLX README","url":"https://github.com/Cthomasdesign/ComfyUI-Krea2-MLX/blob/98cb8b5380256e598b6e4f7886b97fb6141fd5fa/README.md","checkedAt":"2026-10-24","evidence":"> **macOS on Apple Silicon (M-series) only.**"},{"label":"ComfyUI-Krea2-MLX README, resident model","url":"https://github.com/Cthomasdesign/ComfyUI-Krea2-MLX/blob/98cb8b5380256e598b6e4f7886b97fb6141fd5fa/README.md","checkedAt":"2026-10-24","evidence":"with encoder/VAE) stays resident."}]},{"id":"diffusers","name":"Diffusers","memoryPolicy":"resident-sum","formats":["safetensors","gguf"],"platforms":["mac","nvidia"],"support":"official","activationBytes":null,"overheadBytes":null,"note":"Runs on the PyTorch mps device on Mac and CUDA on NVIDIA. GGUF loads through from_single_file on the model class, not the pipeline. Resident size is the sum of the weights, an upper bound.","sources":[{"label":"Diffusers MPS guide","url":"https://github.com/huggingface/diffusers/blob/aaf57d03a3edfbd9de457d2872417a92f149a91d/docs/source/en/optimization/mps.md","checkedAt":"2026-10-24","evidence":"Diffusers is compatible with Apple silicon (M1/M2 chips) using the PyTorch"},{"label":"Diffusers GGUF guide","url":"https://github.com/huggingface/diffusers/blob/aaf57d03a3edfbd9de457d2872417a92f149a91d/docs/source/en/quantization/gguf.md","checkedAt":"2026-10-24","evidence":"Diffusers supports loading checkpoints prequantized and saved in the GGUF format via `from_single_file` loading with Model classes."},{"label":"Diffusers GGUF guide, CUDA kernels","url":"https://github.com/huggingface/diffusers/blob/aaf57d03a3edfbd9de457d2872417a92f149a91d/docs/source/en/quantization/gguf.md","checkedAt":"2026-10-24","evidence":"Optimized CUDA kernels can accelerate GGUF quantized model inference by approximately 10%."}]},{"id":"diffusers-staged","name":"Diffusers with staged loading","memoryPolicy":"stage-max","formats":["safetensors"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"note":"Loads one large model at a time (text encoder, then transformer, then VAE) with the stageload library. In the same test, eager loading and model CPU offload both went into swap on a 48 GB Mac.","sources":[{"label":"How to run Qwen-Image-2.1 on a 48 GB Mac without swap","url":"https://dev.to/nefayran/how-to-run-qwen-image-21-on-a-48-gb-mac-without-swap-3g4b","checkedAt":"2026-10-24","evidence":"Staged loading, with one of the two large models in memory at a time, finished without swap."}]},{"id":"fastvideo","name":"FastVideo MLX (FastMetal)","memoryPolicy":"resident-sum","formats":["mlx"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"note":"FastMetal checkpoints are distilled INT8 derivatives; their times are not base-model figures.","sources":[{"label":"FastVideo MLX install guide","url":"https://raw.githubusercontent.com/hao-ai-lab/FastVideo/main/docs/getting_started/installation/mlx.md","checkedAt":"2026-10-24","evidence":"- macOS 14 or newer"},{"label":"FastVideo MLX install guide","url":"https://raw.githubusercontent.com/hao-ai-lab/FastVideo/main/docs/getting_started/installation/mlx.md","checkedAt":"2026-10-24","evidence":"FastMetal 1.3B and 5B: 16 GB unified memory and up"}]},{"id":"flux-2-swift-mlx","name":"flux-2-swift-mlx","memoryPolicy":"resident-sum","formats":["safetensors","mlx"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"note":"Swift and MLX app and CLI for FLUX.2. It loads the BF16 transformer from the Black Forest Labs repository and takes the text encoder from separate MLX repositories. Its published peak covers transformer, VAE and activations.","sources":[{"label":"flux-2-swift-mlx README","url":"https://github.com/VincentGourbin/flux-2-swift-mlx/blob/15f344e8c1f3e48e80a927a832dbe7343bb5babc/README.md","checkedAt":"2026-10-24","evidence":"The models are downloaded automatically from HuggingFace on first run."},{"label":"flux-2-swift-mlx model registry","url":"https://github.com/VincentGourbin/flux-2-swift-mlx/blob/15f344e8c1f3e48e80a927a832dbe7343bb5babc/Sources/Flux2Core/Configuration/ModelRegistry.swift","checkedAt":"2026-10-24","evidence":"return \"black-forest-labs/FLUX.2-klein-4B\""}]},{"id":"h3c","name":"antirez/h3.c (Metal)","memoryPolicy":"ssd-streaming","formats":["safetensors"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"note":"The ssd-streaming policy applies only with the --ssd-streaming flag; the default keeps the DiT resident.","sources":[{"label":"h3.c README","url":"https://raw.githubusercontent.com/antirez/h3.c/8974cc055ea9c02fcd14cc27dfda3e1027c05153/README.md","checkedAt":"2026-10-24","evidence":"keeps two DiT blocks in memory and reads the next block from SSD while the GPU"},{"label":"h3.c README","url":"https://raw.githubusercontent.com/antirez/h3.c/8974cc055ea9c02fcd14cc27dfda3e1027c05153/README.md","checkedAt":"2026-10-24","evidence":"SSD streaming is an explicit memory/speed tradeoff and is not the default."}]},{"id":"ltx-pipelines","name":"Lightricks/LTX-2 ltx-pipelines","memoryPolicy":"resident-sum","formats":["safetensors"],"platforms":["mac","nvidia"],"support":"official","activationBytes":null,"overheadBytes":null,"note":"No Lightricks Mac benchmark was found.","sources":[{"label":"LTX-2 README","url":"https://github.com/Lightricks/LTX-2/blob/main/README.md","checkedAt":"2026-10-24","evidence":"on Windows and macOS it is skipped automatically and decoding falls back to a Triton or eager implementation"}]},{"id":"ltx-video-mlx","name":"appautomaton/ltx-video-mlx","memoryPolicy":"resident-sum","formats":["safetensors"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"sources":[{"label":"appautomaton ltx-video-mlx README","url":"https://raw.githubusercontent.com/appautomaton/ltx-video-mlx/83b4881d70304013231ce03509227af9c0df500c/README.md","checkedAt":"2026-10-24","evidence":"8-bit / 4-bit quantized inference via MLX"}]},{"id":"mflux","name":"mflux","version":"0.23.0","memoryPolicy":"resident-sum","formats":["safetensors"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"install":{"label":"Install with uv","command":"uv tool install --upgrade mflux"},"note":"MLX-native runner. It reads the original Hugging Face weights and quantizes them while loading (--quantize 3, 4, 5, 6 or 8). --low-ram lowers memory use at a performance cost.","sources":[{"label":"mflux README, install","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/README.md","checkedAt":"2026-10-24","evidence":"uv tool install --upgrade mflux"},{"label":"mflux pyproject","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/pyproject.toml","checkedAt":"2026-10-24","evidence":"version = \"0.23.0\""},{"label":"mflux common README, quantization","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/src/mflux/models/common/README.md","checkedAt":"2026-10-24","evidence":"Most models support `--quantize` (3, 4, 5, 6, 8). This is the on-the-fly option, and quantization is performed as weights are loaded."},{"label":"mflux common README, low RAM","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/src/mflux/models/common/README.md","checkedAt":"2026-10-24","evidence":"Use `--low-ram` to reduce memory usage (at the cost of performance)."}]},{"id":"mlx-serve","name":"ddalcu/mlx-serve","memoryPolicy":"resident-sum","formats":["mlx"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"sources":[{"label":"MLX Serve 4-bit pack card","url":"https://huggingface.co/ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit","checkedAt":"2026-10-24","evidence":"running natively on Apple"}]},{"id":"mlx-video","name":"Blaizzy/mlx-video","memoryPolicy":"resident-sum","formats":["mlx","safetensors"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"sources":[{"label":"mlx-video Wan README","url":"https://github.com/Blaizzy/mlx-video/blob/main/mlx_video/models/wan_2/README.md","checkedAt":"2026-10-24","evidence":"| **Sizes** | 1.3B, 14B | 14B | 14B | 5B |"}]},{"id":"pocketai-mlx","name":"PocketAI Wan2.2 MLX (pocketai-wan-video)","memoryPolicy":"resident-sum","formats":["mlx"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"sources":[{"label":"PocketAI Wan2.2 T2V A14B MLX card","url":"https://huggingface.co/PocketAiHub/Wan2.2-T2V-A14B-MLX","checkedAt":"2026-10-24","evidence":"Apple-Silicon-native Q4 MLX packaging of Wan2.2 T2V A14B"}]},{"id":"rapid-mlx","name":"Rapid-MLX","memoryPolicy":"resident-sum","formats":["safetensors","mlx"],"platforms":["mac"],"support":"community","activationBytes":null,"overheadBytes":null,"note":"MLX-native image server with an OpenAI-compatible images endpoint. SDXL Base quantizes the UNet to 8-bit at load; SD 3.5 Large uses a 4-bit MLX checkpoint.","sources":[{"label":"Rapid-MLX SDXL Base dogfood","url":"https://github.com/raullenchai/Rapid-MLX/blob/e8755711ffedc1365409c89909b6f958be51cded/docs/engineering/performance/2026-09-05-sdxl-base-dogfood.md","checkedAt":"2026-10-24","evidence":"completed real MLX-native text-to-image generations"},{"label":"Rapid-MLX SD 3.5 Large dogfood","url":"https://github.com/raullenchai/Rapid-MLX/blob/dc60da43b612d70ffe29c5b6c6572fceb3cd67ab/docs/engineering/performance/2026-09-05-sd35-large-dogfood.md","checkedAt":"2026-10-24","evidence":"completed real MLX-native text-to-image generation"}]},{"id":"stable-diffusion-cpp","name":"stable-diffusion.cpp","memoryPolicy":"resident-sum","formats":["gguf","safetensors"],"platforms":["mac","nvidia"],"support":"official","activationBytes":null,"overheadBytes":null,"note":"C/C++ runner with Metal on Mac and CUDA on NVIDIA. It reads GGUF and safetensors, and its docs keep a page per model family. Resident size is the sum of the weights, an upper bound.","sources":[{"label":"stable-diffusion.cpp README, backends","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"  - CUDA\n  - Vulkan\n  - Metal"},{"label":"stable-diffusion.cpp README, formats","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"  - Safetensors (`.safetensors`)\n  - GGUF (`.gguf`)"}]}],"models":[{"id":"flux-2-dev","name":"FLUX.2 [dev]","modality":"image","creator":"Black Forest Labs","url":"https://modelfit.io/image/flux-2-dev/","officialRepository":"https://huggingface.co/black-forest-labs/FLUX.2-dev","license":{"name":"FLUX Non-Commercial License","url":"https://huggingface.co/black-forest-labs/FLUX.2-dev/resolve/main/LICENSE.md","commercial":"restricted","excludedTerritories":[],"note":"Model use is non-commercial. Output may be used commercially except as expressly prohibited, and may not train, fine-tune or distill a competing FLUX model."},"components":[{"id":"dit","name":"FLUX.2 transformer","role":"dit","stage":"denoise","required":true,"sources":[{"label":"FLUX.2-dev model card","url":"https://huggingface.co/black-forest-labs/FLUX.2-dev/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"`FLUX.2 [dev]` is a 32 billion parameter rectified flow transformer capable of generating, editing and combining images based on text instructions."}],"parametersB":32},{"id":"text_encoder","name":"Mistral Small 3.2 text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"FLUX.2 reference README","url":"https://github.com/black-forest-labs/flux2/blob/main/README.md","checkedAt":"2026-10-24","evidence":"Mistral-Small-3.2-24B-Instruct-2506"}],"parametersB":24},{"id":"vae","name":"FLUX.2 VAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"FLUX.2-dev model card","url":"https://huggingface.co/black-forest-labs/FLUX.2-dev/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"VAE still in bf16"}]}],"builds":[{"id":"bf16","label":"BF16 safetensors","format":"safetensors","quantization":"BF16","apps":["flux-2-swift-mlx","diffusers"],"files":[{"repo":"black-forest-labs/FLUX.2-dev","path":"transformer/diffusion_pytorch_model-00001-of-00007.safetensors","componentId":"dit","bytes":9935797200,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"transformer/diffusion_pytorch_model-00002-of-00007.safetensors","componentId":"dit","bytes":9890181048,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"transformer/diffusion_pytorch_model-00003-of-00007.safetensors","componentId":"dit","bytes":9814681480,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"transformer/diffusion_pytorch_model-00004-of-00007.safetensors","componentId":"dit","bytes":9814681536,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"transformer/diffusion_pytorch_model-00005-of-00007.safetensors","componentId":"dit","bytes":9814681536,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"transformer/diffusion_pytorch_model-00006-of-00007.safetensors","componentId":"dit","bytes":9814681536,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"transformer/diffusion_pytorch_model-00007-of-00007.safetensors","componentId":"dit","bytes":5361898792,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00001-of-00010.safetensors","componentId":"text_encoder","bytes":4883550696,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00002-of-00010.safetensors","componentId":"text_encoder","bytes":4781593336,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00003-of-00010.safetensors","componentId":"text_encoder","bytes":4886472224,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00004-of-00010.safetensors","componentId":"text_encoder","bytes":4781593376,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00005-of-00010.safetensors","componentId":"text_encoder","bytes":4781593368,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00006-of-00010.safetensors","componentId":"text_encoder","bytes":4886472248,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00007-of-00010.safetensors","componentId":"text_encoder","bytes":4781593376,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00008-of-00010.safetensors","componentId":"text_encoder","bytes":4781593368,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00009-of-00010.safetensors","componentId":"text_encoder","bytes":4886472248,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"text_encoder/model-00010-of-00010.safetensors","componentId":"text_encoder","bytes":4571866320,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":336213556,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"}],"sources":[{"label":"flux-2-swift-mlx model registry","url":"https://github.com/VincentGourbin/flux-2-swift-mlx/blob/15f344e8c1f3e48e80a927a832dbe7343bb5babc/Sources/Flux2Core/Configuration/ModelRegistry.swift","checkedAt":"2026-10-24","evidence":"return \"black-forest-labs/FLUX.2-dev\""},{"label":"FLUX.2-dev model card","url":"https://huggingface.co/black-forest-labs/FLUX.2-dev/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"from diffusers import Flux2Pipeline"}]},{"id":"gguf-q8_0","label":"Q8_0 GGUF","format":"gguf","quantization":"Q8_0","apps":["comfyui","stable-diffusion-cpp"],"files":[{"repo":"city96/FLUX.2-dev-gguf","path":"flux2-dev-Q8_0.gguf","componentId":"dit","bytes":35002602464,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/city96/FLUX.2-dev-gguf?blobs=true"},{"repo":"unsloth/Mistral-Small-3.2-24B-Instruct-2506-GGUF","path":"Mistral-Small-3.2-24B-Instruct-2506-Q4_K_M.gguf","componentId":"text_encoder","bytes":14333922848,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/Mistral-Small-3.2-24B-Instruct-2506-GGUF?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"ae.safetensors","componentId":"vae","bytes":336211292,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"}],"sources":[{"label":"FLUX.2-dev GGUF card","url":"https://huggingface.co/city96/FLUX.2-dev-gguf","checkedAt":"2026-10-24","evidence":"The model files can be used in [ComfyUI](https://github.com/comfyanonymous/ComfyUI/) with the [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF) custom node."},{"label":"stable-diffusion.cpp FLUX.2 guide","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/docs/flux2.md","checkedAt":"2026-10-24","evidence":"- Download FLUX.2-dev\n    - gguf: https://huggingface.co/city96/FLUX.2-dev-gguf/tree/main"}]},{"id":"gguf-q4_k_m","label":"Q4_K_M GGUF","format":"gguf","quantization":"Q4_K_M","apps":["comfyui","stable-diffusion-cpp"],"files":[{"repo":"city96/FLUX.2-dev-gguf","path":"flux2-dev-Q4_K_M.gguf","componentId":"dit","bytes":20082414560,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/city96/FLUX.2-dev-gguf?blobs=true"},{"repo":"unsloth/Mistral-Small-3.2-24B-Instruct-2506-GGUF","path":"Mistral-Small-3.2-24B-Instruct-2506-Q4_K_M.gguf","componentId":"text_encoder","bytes":14333922848,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/Mistral-Small-3.2-24B-Instruct-2506-GGUF?blobs=true"},{"repo":"black-forest-labs/FLUX.2-dev","path":"ae.safetensors","componentId":"vae","bytes":336211292,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-dev?blobs=true"}],"sources":[{"label":"FLUX.2-dev GGUF card","url":"https://huggingface.co/city96/FLUX.2-dev-gguf","checkedAt":"2026-10-24","evidence":"The model files can be used in [ComfyUI](https://github.com/comfyanonymous/ComfyUI/) with the [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF) custom node."},{"label":"stable-diffusion.cpp FLUX.2 guide","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/docs/flux2.md","checkedAt":"2026-10-24","evidence":"- Download FLUX.2-dev\n    - gguf: https://huggingface.co/city96/FLUX.2-dev-gguf/tree/main"}]}],"publishedMeasurements":[{"id":"flux-2-dev-swift-mlx-bf16-m2-ultra-96gb-1024","modelId":"flux-2-dev","buildId":"bf16","appId":"flux-2-swift-mlx","machine":{"chip":"Apple M2 Ultra","ramGb":96},"workload":{"resolution":"1024x1024","steps":28,"quantization":"BF16","notes":"Dev 32B, seed 42, macOS Tahoe 26.2, MLX Swift 0.30.2. Inference time only. Peak is transformer, VAE and activations, not whole-process memory."},"outcome":"success","timeSeconds":1758.6,"peakMemoryGb":71.93,"peakMemoryText":"71 930 MB","publishedAt":"2026-02-11","grade":"A","source":{"label":"flux-2-swift-mlx quantization benchmark","url":"https://github.com/VincentGourbin/flux-2-swift-mlx/blob/15f344e8c1f3e48e80a927a832dbe7343bb5babc/docs/examples/quantization-benchmark/README.md","checkedAt":"2026-10-24","evidence":"Measured on **Apple M2 Ultra 96GB**, macOS Tahoe 26.2, MLX Swift 0.30.2."}}],"quality":{"modelId":"flux-2-dev","category":"text_to_image","sourceName":"flux-2-dev","rank":38,"total":82,"votes":77988,"rating":1145.7703186822487,"low":1141.6232865632555,"high":1149.9173508012418,"snapshot":"2026-10-07","attribution":"Blind human votes from Arena (https://arena.ai), CC BY 4.0. Arena checkpoint, not a quantized local-build score.","dataset":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset"},"sources":[{"label":"FLUX.2-dev license","url":"https://huggingface.co/black-forest-labs/FLUX.2-dev/resolve/main/LICENSE.md","checkedAt":"2026-10-24","evidence":"You may only access, use, Distribute, or create Derivatives of the FLUX Model or Derivatives for Non-Commercial Purposes."},{"label":"FLUX.2-dev license, outputs","url":"https://huggingface.co/black-forest-labs/FLUX.2-dev/resolve/main/LICENSE.md","checkedAt":"2026-10-24","evidence":"You may use Output for any purpose (including for commercial purposes), except as expressly prohibited herein."},{"label":"Draw Things App Store listing data","url":"https://itunes.apple.com/lookup?id=6444050820&country=us","checkedAt":"2026-10-24","evidence":"Harness the latest models like SDXL, FLUX.1, Wan 2.2, Qwen Image, Z Image, FLUX.2 and LTX-2.3"}]},{"id":"flux-2-klein","name":"FLUX.2 [klein]","modality":"image","creator":"Black Forest Labs","url":"https://modelfit.io/image/flux-2-klein/","officialRepository":"https://huggingface.co/black-forest-labs/FLUX.2-klein-4B","license":{"name":"Apache 2.0 (4B), FLUX Non-Commercial License (9B)","url":"https://huggingface.co/black-forest-labs/FLUX.2-klein-4B","commercial":"restricted","excludedTerritories":[],"note":"The 4B model allows commercial use under Apache 2.0. Model use of the 9B models is non-commercial, and filters or manual review are required."},"components":[{"id":"dit","name":"FLUX.2 klein transformer","role":"dit","stage":"denoise","required":true,"sources":[{"label":"FLUX.2 klein 4B model card","url":"https://huggingface.co/black-forest-labs/FLUX.2-klein-4B","checkedAt":"2026-10-24","evidence":"FLUX.2 [klein] 4B is a 4 billion parameter rectified flow transformer"}]},{"id":"text_encoder","name":"Qwen3 text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"FLUX.2 reference code","url":"https://github.com/black-forest-labs/flux2/blob/main/src/flux2/text_encoder.py","checkedAt":"2026-10-24","evidence":"class Qwen3Embedder(nn.Module):"}]},{"id":"vae","name":"FLUX.2 VAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"FLUX.2 reference code","url":"https://github.com/black-forest-labs/flux2/blob/main/src/flux2/autoencoder.py","checkedAt":"2026-10-24","evidence":"class AutoEncoderParams:"}]}],"builds":[{"id":"klein-4b-bf16","label":"Klein 4B BF16","format":"safetensors","quantization":"BF16","apps":["flux-2-swift-mlx","diffusers"],"files":[{"repo":"black-forest-labs/FLUX.2-klein-4B","path":"flux-2-klein-4b.safetensors","componentId":"dit","bytes":7751105712,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-4B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-4B","path":"text_encoder/model-00001-of-00002.safetensors","componentId":"text_encoder","bytes":4967215360,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-4B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-4B","path":"text_encoder/model-00002-of-00002.safetensors","componentId":"text_encoder","bytes":3077766632,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-4B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-4B","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":168120878,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-4B?blobs=true"}],"sources":[{"label":"flux-2-swift-mlx model registry","url":"https://github.com/VincentGourbin/flux-2-swift-mlx/blob/15f344e8c1f3e48e80a927a832dbe7343bb5babc/Sources/Flux2Core/Configuration/ModelRegistry.swift","checkedAt":"2026-10-24","evidence":"return \"black-forest-labs/FLUX.2-klein-4B\""},{"label":"FLUX.2 klein 4B model card","url":"https://huggingface.co/black-forest-labs/FLUX.2-klein-4B","checkedAt":"2026-10-24","evidence":"from diffusers import Flux2KleinPipeline"}]},{"id":"klein-9b-bf16","label":"Klein 9B BF16","format":"safetensors","quantization":"BF16","apps":["flux-2-swift-mlx","diffusers"],"files":[{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"flux-2-klein-9b.safetensors","componentId":"dit","bytes":18157185168,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"text_encoder/model-00001-of-00004.safetensors","componentId":"text_encoder","bytes":4902257696,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"text_encoder/model-00002-of-00004.safetensors","componentId":"text_encoder","bytes":4915960368,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"text_encoder/model-00003-of-00004.safetensors","componentId":"text_encoder","bytes":4983068496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"text_encoder/model-00004-of-00004.safetensors","componentId":"text_encoder","bytes":1580230264,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":168120878,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"}],"sources":[{"label":"flux-2-swift-mlx model registry","url":"https://github.com/VincentGourbin/flux-2-swift-mlx/blob/15f344e8c1f3e48e80a927a832dbe7343bb5babc/Sources/Flux2Core/Configuration/ModelRegistry.swift","checkedAt":"2026-10-24","evidence":"return \"black-forest-labs/FLUX.2-klein-9B\""},{"label":"FLUX.2 klein 9B model card","url":"https://huggingface.co/black-forest-labs/FLUX.2-klein-9B/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"from diffusers import Flux2KleinPipeline"}]},{"id":"klein-4b-gguf-q8_0","label":"Klein 4B Q8_0 GGUF","format":"gguf","quantization":"Q8_0","apps":["stable-diffusion-cpp"],"files":[{"repo":"unsloth/FLUX.2-klein-4B-GGUF","path":"flux-2-klein-4b-Q8_0.gguf","componentId":"dit","bytes":4300644928,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/FLUX.2-klein-4B-GGUF?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-4B","path":"text_encoder/model-00001-of-00002.safetensors","componentId":"text_encoder","bytes":4967215360,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-4B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-4B","path":"text_encoder/model-00002-of-00002.safetensors","componentId":"text_encoder","bytes":3077766632,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-4B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-4B","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":168120878,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-4B?blobs=true"}],"sources":[{"label":"FLUX.2 klein GGUF card","url":"https://huggingface.co/unsloth/FLUX.2-klein-4B-GGUF","checkedAt":"2026-10-24","evidence":"Uses tooling from [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF) by city96."},{"label":"stable-diffusion.cpp README","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"[FLUX.2-dev/FLUX.2-klein](./docs/flux2.md)"}]},{"id":"klein-9b-gguf-q4_k_m","label":"Klein 9B Q4_K_M GGUF","format":"gguf","quantization":"Q4_K_M","apps":["stable-diffusion-cpp"],"files":[{"repo":"unsloth/FLUX.2-klein-9B-GGUF","path":"flux-2-klein-9b-Q4_K_M.gguf","componentId":"dit","bytes":5909829920,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/FLUX.2-klein-9B-GGUF?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"text_encoder/model-00001-of-00004.safetensors","componentId":"text_encoder","bytes":4902257696,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"text_encoder/model-00002-of-00004.safetensors","componentId":"text_encoder","bytes":4915960368,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"text_encoder/model-00003-of-00004.safetensors","componentId":"text_encoder","bytes":4983068496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"text_encoder/model-00004-of-00004.safetensors","componentId":"text_encoder","bytes":1580230264,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"},{"repo":"black-forest-labs/FLUX.2-klein-9B","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":168120878,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/FLUX.2-klein-9B?blobs=true"}],"sources":[{"label":"FLUX.2 klein GGUF card","url":"https://huggingface.co/unsloth/FLUX.2-klein-9B-GGUF","checkedAt":"2026-10-24","evidence":"Uses tooling from [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF) by city96."},{"label":"stable-diffusion.cpp README","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"[FLUX.2-dev/FLUX.2-klein](./docs/flux2.md)"}]}],"publishedMeasurements":[{"id":"flux-2-klein-4b-swift-mlx-bf16-m2-ultra-96gb-1024","modelId":"flux-2-klein","buildId":"klein-4b-bf16","appId":"flux-2-swift-mlx","machine":{"chip":"Apple M2 Ultra","ramGb":96},"workload":{"resolution":"1024x1024","steps":4,"quantization":"BF16","notes":"Klein 4B distilled, seed 42, macOS Tahoe 26.2, MLX Swift 0.30.2. Inference time only. Peak is transformer, VAE and activations, not whole-process memory."},"outcome":"success","timeSeconds":25.6,"peakMemoryGb":17.846,"peakMemoryText":"17 846 MB","publishedAt":"2026-02-11","grade":"A","source":{"label":"flux-2-swift-mlx quantization benchmark","url":"https://github.com/VincentGourbin/flux-2-swift-mlx/blob/15f344e8c1f3e48e80a927a832dbe7343bb5babc/docs/examples/quantization-benchmark/README.md","checkedAt":"2026-10-24","evidence":"Measured on **Apple M2 Ultra 96GB**, macOS Tahoe 26.2, MLX Swift 0.30.2."}},{"id":"flux-2-klein-9b-swift-mlx-bf16-m2-ultra-96gb-1024","modelId":"flux-2-klein","buildId":"klein-9b-bf16","appId":"flux-2-swift-mlx","machine":{"chip":"Apple M2 Ultra","ramGb":96},"workload":{"resolution":"1024x1024","steps":4,"quantization":"BF16","notes":"Klein 9B distilled, seed 42, macOS Tahoe 26.2, MLX Swift 0.30.2. Inference time only. Peak is transformer, VAE and activations, not whole-process memory."},"outcome":"success","timeSeconds":54.9,"peakMemoryGb":27.779,"peakMemoryText":"27 779 MB","publishedAt":"2026-02-11","grade":"A","source":{"label":"flux-2-swift-mlx quantization benchmark","url":"https://github.com/VincentGourbin/flux-2-swift-mlx/blob/15f344e8c1f3e48e80a927a832dbe7343bb5babc/docs/examples/quantization-benchmark/README.md","checkedAt":"2026-10-24","evidence":"Measured on **Apple M2 Ultra 96GB**, macOS Tahoe 26.2, MLX Swift 0.30.2."}}],"quality":null,"sources":[{"label":"FLUX.2 klein 4B model card, license","url":"https://huggingface.co/black-forest-labs/FLUX.2-klein-4B","checkedAt":"2026-10-24","evidence":"FLUX.2 [klein] 4B models under an Apache 2.0 license and the release of the FLUX.2 [klein] 9B models under a non-commercial license"},{"label":"FLUX.2 klein 4B model card, license terms","url":"https://huggingface.co/black-forest-labs/FLUX.2-klein-4B","checkedAt":"2026-10-24","evidence":"Filters or manual review must be used with the FLUX.2 [klein] 9B models under the terms of the FLUX Non-Commercial License"},{"label":"mflux README","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/README.md","checkedAt":"2026-10-24","evidence":"[FLUX.2](src/mflux/models/flux2/README.md)"},{"label":"FLUX.2 klein 4B model card, ComfyUI","url":"https://huggingface.co/black-forest-labs/FLUX.2-klein-4B","checkedAt":"2026-10-24","evidence":"is also available in both [ComfyUI](https://github.com/comfyanonymous/ComfyUI) and [Diffusers]"},{"label":"Draw Things App Store listing data","url":"https://itunes.apple.com/lookup?id=6444050820&country=us","checkedAt":"2026-10-24","evidence":"Harness the latest models like SDXL, FLUX.1, Wan 2.2, Qwen Image, Z Image, FLUX.2 and LTX-2.3"}]},{"id":"flux-3-klein","name":"FLUX 3 Image Klein","modality":"image","creator":"Black Forest Labs","url":"https://modelfit.io/image/flux-3-klein/","officialRepository":"https://huggingface.co/black-forest-labs/flux-3-image-klein","license":{"name":"FLUX Kommunity License v1.0","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein/resolve/main/LICENSE.md","commercial":"restricted","excludedTerritories":[],"note":"Ordinary model use is non-commercial. Output generation has an exception for qualifying users below US$5,000,000 gross annualized revenue."},"components":[{"id":"dit","name":"FLUX 3 transformer","role":"dit","stage":"denoise","required":true,"sources":[{"label":"FLUX 3 Image Klein model card","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"The 5.6B parameter label refers to the image transformer; the text encoder is separate."}],"parametersB":5.6},{"id":"text_encoder","name":"Qwen3-VL-4B-Instruct text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"FLUX 3 Image Klein model card","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"with Qwen3-VL-4B-Instruct as the text encoder."}]},{"id":"vae","name":"FLUX image autoencoder","role":"vae","stage":"decode","required":true,"sources":[{"label":"FLUX 3 Image Klein model card","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"together with the Qwen3-VL-4B-Instruct text encoder and image autoencoder."}]},{"id":"prompt_upsampler","name":"Prompt upsampler (separate 9B model)","role":"other","stage":"text","required":false,"sources":[{"label":"FLUX 3 Image Klein prompt upsampler card","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein-pup/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"Qwen3.5-9B"}]}],"builds":[{"id":"bf16","label":"BF16 safetensors","format":"safetensors","quantization":"BF16","apps":["diffusers"],"files":[{"repo":"black-forest-labs/flux-3-image-klein","path":"transformer/diffusion_pytorch_model-00001-of-00002.safetensors","componentId":"dit","bytes":9907529352,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"transformer/diffusion_pytorch_model-00002-of-00002.safetensors","componentId":"dit","bytes":1189878064,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"text_encoder/model-00001-of-00002.safetensors","componentId":"text_encoder","bytes":4967229296,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"text_encoder/model-00002-of-00002.safetensors","componentId":"text_encoder","bytes":3908490048,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":336213556,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"}],"sources":[{"label":"FLUX 3 Image Klein model card","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"FLUX 3 Image Klein uses a modular pipeline."}]},{"id":"mflux-4bit","label":"mflux 4-bit","format":"safetensors","quantization":"4-bit (mflux --quantize 4)","apps":["mflux"],"files":[{"repo":"black-forest-labs/flux-3-image-klein","path":"transformer/diffusion_pytorch_model-00001-of-00002.safetensors","componentId":"dit","bytes":9907529352,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"transformer/diffusion_pytorch_model-00002-of-00002.safetensors","componentId":"dit","bytes":1189878064,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"text_encoder/model-00001-of-00002.safetensors","componentId":"text_encoder","bytes":4967229296,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"text_encoder/model-00002-of-00002.safetensors","componentId":"text_encoder","bytes":3908490048,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":336213556,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"}],"sources":[{"label":"mflux model registry","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/src/mflux/models/common/config/model_config.py","checkedAt":"2026-10-24","evidence":"model_name=\"black-forest-labs/flux-3-image-klein\","},{"label":"mflux README","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/README.md","checkedAt":"2026-10-24","evidence":"[FLUX 3 Image Klein](src/mflux/models/flux3/README.md)"}]},{"id":"gguf-q8_0","label":"Q8_0 GGUF","format":"gguf","quantization":"Q8_0","apps":["diffusers"],"files":[{"repo":"unsloth/flux-3-image-klein-GGUF","path":"flux-3-image-klein-Q8_0.gguf","componentId":"dit","bytes":6092755872,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/flux-3-image-klein-GGUF?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"text_encoder/model-00001-of-00002.safetensors","componentId":"text_encoder","bytes":4967229296,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"text_encoder/model-00002-of-00002.safetensors","componentId":"text_encoder","bytes":3908490048,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":336213556,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"}],"sources":[{"label":"FLUX 3 Image Klein GGUF card","url":"https://huggingface.co/unsloth/flux-3-image-klein-GGUF","checkedAt":"2026-10-24","evidence":"Loads in diffusers through `Flux3Transformer2DModel.from_single_file(..., quantization_config=GGUFQuantizationConfig(...))`."},{"label":"Diffusers GGUF guide","url":"https://github.com/huggingface/diffusers/blob/aaf57d03a3edfbd9de457d2872417a92f149a91d/docs/source/en/quantization/gguf.md","checkedAt":"2026-10-24","evidence":"Diffusers supports loading checkpoints prequantized and saved in the GGUF format via `from_single_file` loading with Model classes."}]},{"id":"gguf-q4_k_m","label":"Q4_K_M GGUF","format":"gguf","quantization":"Q4_K_M","apps":["diffusers"],"files":[{"repo":"unsloth/flux-3-image-klein-GGUF","path":"flux-3-image-klein-Q4_K_M.gguf","componentId":"dit","bytes":3666908064,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/flux-3-image-klein-GGUF?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"text_encoder/model-00001-of-00002.safetensors","componentId":"text_encoder","bytes":4967229296,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"text_encoder/model-00002-of-00002.safetensors","componentId":"text_encoder","bytes":3908490048,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"},{"repo":"black-forest-labs/flux-3-image-klein","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":336213556,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/black-forest-labs/flux-3-image-klein?blobs=true"}],"sources":[{"label":"FLUX 3 Image Klein GGUF card","url":"https://huggingface.co/unsloth/flux-3-image-klein-GGUF","checkedAt":"2026-10-24","evidence":"Loads in diffusers through `Flux3Transformer2DModel.from_single_file(..., quantization_config=GGUFQuantizationConfig(...))`."},{"label":"Diffusers GGUF guide","url":"https://github.com/huggingface/diffusers/blob/aaf57d03a3edfbd9de457d2872417a92f149a91d/docs/source/en/quantization/gguf.md","checkedAt":"2026-10-24","evidence":"Diffusers supports loading checkpoints prequantized and saved in the GGUF format via `from_single_file` loading with Model classes."}]}],"publishedMeasurements":[{"id":"flux-3-klein-mflux-4bit-default-1-m1-max-32gb-512","modelId":"flux-3-klein","buildId":"mflux-4bit","appId":"mflux","machine":{"chip":"Apple M1 Max","ramGb":32},"workload":{"resolution":"512x512","steps":16,"quantization":"4-bit (mflux --quantize 4)","notes":"Default mode. Seed 42, mflux 0.23.0, MLX 0.32.0, macOS 14.6, local 4-bit conversion of the distilled checkpoint. External wall time and peak MLX allocation. Other applications were active."},"outcome":"success","timeSeconds":49.45,"peakMemoryGb":6.71,"publishedAt":"2026-10-20","grade":"A","source":{"label":"mflux FLUX 3 memory benchmark","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/scripts/flux3/MEMORY_BENCHMARK.md","checkedAt":"2026-10-24","evidence":"## M1 Max 32 GB results, 20 October 2026"}},{"id":"flux-3-klein-mflux-4bit-lowram-1-m1-max-32gb-512","modelId":"flux-3-klein","buildId":"mflux-4bit","appId":"mflux","machine":{"chip":"Apple M1 Max","ramGb":32},"workload":{"resolution":"512x512","steps":16,"quantization":"4-bit (mflux --quantize 4)","notes":"--low-ram mode. Seed 42, mflux 0.23.0, MLX 0.32.0, macOS 14.6, local 4-bit conversion of the distilled checkpoint. External wall time and peak MLX allocation. Other applications were active."},"outcome":"success","timeSeconds":52.63,"peakMemoryGb":3.997,"publishedAt":"2026-10-20","grade":"A","source":{"label":"mflux FLUX 3 memory benchmark","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/scripts/flux3/MEMORY_BENCHMARK.md","checkedAt":"2026-10-24","evidence":"## M1 Max 32 GB results, 20 October 2026"}}],"quality":null,"sources":[{"label":"FLUX 3 Image Klein license","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein/resolve/main/LICENSE.md","checkedAt":"2026-10-24","evidence":"gross annualized revenue is less than US$5,000,000"},{"label":"stable-diffusion.cpp README","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"[FLUX.3 Image Klein / Klein Base](./docs/flux3.md)"}]},{"id":"krea-2","name":"Krea 2 Turbo","modality":"image","creator":"Krea","url":"https://modelfit.io/image/krea-2/","officialRepository":"https://huggingface.co/krea/Krea-2-Turbo","license":{"name":"Krea 2 Community License Agreement v.1","url":"https://github.com/krea-ai/krea-2/blob/db3984fbc6e13b34c0064990fc2d95ac64d00058/docs/KREA-2-COMMUNITY-LICENSE","commercial":"restricted","excludedTerritories":[],"note":"Commercial use needs total company-wide annual revenue below USD $1,000,000 on a trailing twelve-month basis, or an Enterprise License."},"components":[{"id":"dit","name":"Krea 2 transformer","role":"dit","stage":"denoise","required":true,"sources":[{"label":"Krea 2 inference code","url":"https://github.com/krea-ai/krea-2/blob/db3984fbc6e13b34c0064990fc2d95ac64d00058/inference.py","checkedAt":"2026-10-24","evidence":"SingleStreamDiT"}]},{"id":"text_encoder","name":"Qwen3-VL-4B-Instruct text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"Krea 2 inference code","url":"https://github.com/krea-ai/krea-2/blob/db3984fbc6e13b34c0064990fc2d95ac64d00058/inference.py","checkedAt":"2026-10-24","evidence":"qwen3_vl_4b = TextEncoderConfig(model_id=\"Qwen/Qwen3-VL-4B-Instruct\")"}]},{"id":"vae","name":"Qwen-Image VAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"Krea 2 autoencoder code","url":"https://github.com/krea-ai/krea-2/blob/db3984fbc6e13b34c0064990fc2d95ac64d00058/autoencoder.py","checkedAt":"2026-10-24","evidence":"qwen-ae-f8-16c: the Qwen-Image VAE (f8, 16 latent channels)."}]}],"builds":[{"id":"turbo-mlx-mixed-4-8","label":"Turbo MLX mixed-4/8","format":"mlx","quantization":"mixed-4/8","apps":["comfyui-krea2-mlx"],"files":[{"repo":"avlp12/Krea-2-Turbo-Alis-MLX-mixed-4-8","path":"transformer_mixed_4_8.safetensors","componentId":"dit","bytes":9840816670,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/avlp12/Krea-2-Turbo-Alis-MLX-mixed-4-8?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"text_encoder/model.safetensors","componentId":"text_encoder","bytes":8875715136,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":507591892,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"}],"sources":[{"label":"ComfyUI-Krea2-MLX README","url":"https://github.com/Cthomasdesign/ComfyUI-Krea2-MLX/blob/98cb8b5380256e598b6e4f7886b97fb6141fd5fa/README.md","checkedAt":"2026-10-24","evidence":"| mixed-4/8 | 9.8 GB |"},{"label":"ComfyUI-Krea2-MLX README, encoder and VAE","url":"https://github.com/Cthomasdesign/ComfyUI-Krea2-MLX/blob/98cb8b5380256e598b6e4f7886b97fb6141fd5fa/README.md","checkedAt":"2026-10-24","evidence":"On first generation the **VAE + Qwen3-VL-4B text encoder** (~9 GB) download automatically"}]},{"id":"turbo-original","label":"Turbo original safetensors","format":"safetensors","quantization":"Original weights","apps":["diffusers"],"files":[{"repo":"krea/Krea-2-Turbo","path":"transformer/diffusion_pytorch_model-00001-of-00003.safetensors","componentId":"dit","bytes":9970652128,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"transformer/diffusion_pytorch_model-00002-of-00003.safetensors","componentId":"dit","bytes":9829444456,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"transformer/diffusion_pytorch_model-00003-of-00003.safetensors","componentId":"dit","bytes":6483240784,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"text_encoder/model.safetensors","componentId":"text_encoder","bytes":8875715136,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":507591892,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"}],"sources":[{"label":"Krea 2 Turbo model card","url":"https://huggingface.co/krea/Krea-2-Turbo/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"from diffusers import Krea2Pipeline"}]},{"id":"turbo-gguf-q8_0","label":"Turbo Q8_0 GGUF","format":"gguf","quantization":"Q8_0","apps":["stable-diffusion-cpp"],"files":[{"repo":"vantagewithai/Krea-2-Turbo-GGUF","path":"krea2_turbo-Q8_0.gguf","componentId":"dit","bytes":13705958688,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/vantagewithai/Krea-2-Turbo-GGUF?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"text_encoder/model.safetensors","componentId":"text_encoder","bytes":8875715136,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":507591892,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"}],"sources":[{"label":"Krea 2 Turbo GGUF card","url":"https://huggingface.co/vantagewithai/Krea-2-Turbo-GGUF","checkedAt":"2026-10-24","evidence":"This repository provides **GGUF quantized version** of the **Krea 2 OSS Turbo**."},{"label":"stable-diffusion.cpp README","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"[Krea2](./docs/krea2.md)"}]},{"id":"turbo-gguf-q4_k_m","label":"Turbo Q4_K_M GGUF","format":"gguf","quantization":"Q4_K_M","apps":["stable-diffusion-cpp"],"files":[{"repo":"vantagewithai/Krea-2-Turbo-GGUF","path":"krea2_turbo-Q4_K_M.gguf","componentId":"dit","bytes":7486289184,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/vantagewithai/Krea-2-Turbo-GGUF?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"text_encoder/model.safetensors","componentId":"text_encoder","bytes":8875715136,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"},{"repo":"krea/Krea-2-Turbo","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":507591892,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/krea/Krea-2-Turbo?blobs=true"}],"sources":[{"label":"Krea 2 Turbo GGUF card","url":"https://huggingface.co/vantagewithai/Krea-2-Turbo-GGUF","checkedAt":"2026-10-24","evidence":"This repository provides **GGUF quantized version** of the **Krea 2 OSS Turbo**."},{"label":"stable-diffusion.cpp README","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"[Krea2](./docs/krea2.md)"}]}],"publishedMeasurements":[{"id":"krea-2-turbo-mlx-mixed-4-8-m1-max-32gb-1024","modelId":"krea-2","buildId":"turbo-mlx-mixed-4-8","appId":"comfyui-krea2-mlx","machine":{"chip":"Apple M1 Max","ramGb":32},"workload":{"resolution":"1024x1024","steps":8,"quantization":"mixed-4/8","notes":"ComfyUI-Krea2-MLX v0.4.0 with the Turbo mixed-4/8 build. Sampler, guidance and run date are not stated; the date is the README commit date."},"outcome":"success","timeSeconds":263,"timeText":"~263s","publishedAt":"2026-07-08","grade":"A","source":{"label":"ComfyUI-Krea2-MLX README","url":"https://github.com/Cthomasdesign/ComfyUI-Krea2-MLX/blob/98cb8b5380256e598b6e4f7886b97fb6141fd5fa/README.md","checkedAt":"2026-10-24","evidence":"measured on an M1 Max 32 GB (mixed-4/8 build, 8 steps):"}}],"quality":null,"sources":[{"label":"Krea 2 Community License","url":"https://github.com/krea-ai/krea-2/blob/db3984fbc6e13b34c0064990fc2d95ac64d00058/docs/KREA-2-COMMUNITY-LICENSE","checkedAt":"2026-10-24","evidence":"calculated on a trailing twelve-month basis"},{"label":"Krea 2 README, resolution","url":"https://github.com/krea-ai/krea-2/blob/db3984fbc6e13b34c0064990fc2d95ac64d00058/README.md","checkedAt":"2026-10-24","evidence":"The model can generate images from 1k ~ 2k resolution."},{"label":"Krea 2 README, Raw settings","url":"https://github.com/krea-ai/krea-2/blob/db3984fbc6e13b34c0064990fc2d95ac64d00058/README.md","checkedAt":"2026-10-24","evidence":"--checkpoint oss_raw --steps 52 --cfg 3.5"},{"label":"mflux README","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/README.md","checkedAt":"2026-10-24","evidence":"[Krea 2]"}]},{"id":"ltx-2-3","name":"LTX-2.3","modality":"video","creator":"Lightricks","url":"https://modelfit.io/video/ltx-2-3/","officialRepository":"https://huggingface.co/Lightricks/LTX-2.3","license":{"name":"LTX-2 Community License Agreement","url":"https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2","commercial":"restricted","excludedTerritories":[],"note":"Entities with annual revenues of at least $10,000,000 require a paid commercial use license."},"components":[{"id":"dit","name":"LTX-2.3 22B transformer (dev or distilled)","role":"dit","stage":"denoise","required":true,"parametersB":22,"sources":[{"label":"LTX-2.3 model card","url":"https://huggingface.co/Lightricks/LTX-2.3","checkedAt":"2026-10-24","evidence":"| ltx-2.3-22b-dev                    | The full model, flexible and trainable in bf16"}]},{"id":"text_encoder","name":"Gemma 3 12B text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"LTX-2.3 models list","url":"https://github.com/Lightricks/LTX-2/blob/main/MODELS-LTX-2.3.md","checkedAt":"2026-10-24","evidence":"https://huggingface.co/google/gemma-3-12b-it-qat-q4_0-unquantized/tree/main"}]},{"id":"video_vae","name":"Video VAE (bundled in the official checkpoints)","role":"vae","stage":"decode","required":false,"sources":[]},{"id":"audio_vae","name":"Audio VAE (bundled in the official checkpoints)","role":"audio","stage":"decode","required":false,"sources":[]},{"id":"text_connector","name":"Text projection and connectors (bundled in the official checkpoints)","role":"other","stage":"text","required":false,"sources":[]},{"id":"distilled_lora","name":"Distilled LoRA","role":"other","stage":"denoise","required":false,"sources":[]}],"builds":[{"id":"official-bf16","label":"Official dev checkpoint (BF16) with Gemma 3","format":"safetensors","quantization":"BF16","apps":["ltx-pipelines"],"files":[{"repo":"Lightricks/LTX-2.3","path":"ltx-2.3-22b-dev.safetensors","componentId":"dit","bytes":46149344974,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Lightricks/LTX-2.3?blobs=true"},{"repo":"google/gemma-3-12b-it-qat-q4_0-unquantized","path":"model-00001-of-00005.safetensors","componentId":"text_encoder","bytes":4979902192,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-it-qat-q4_0-unquantized?blobs=true"},{"repo":"google/gemma-3-12b-it-qat-q4_0-unquantized","path":"model-00002-of-00005.safetensors","componentId":"text_encoder","bytes":4931296592,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-it-qat-q4_0-unquantized?blobs=true"},{"repo":"google/gemma-3-12b-it-qat-q4_0-unquantized","path":"model-00003-of-00005.safetensors","componentId":"text_encoder","bytes":4931296656,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-it-qat-q4_0-unquantized?blobs=true"},{"repo":"google/gemma-3-12b-it-qat-q4_0-unquantized","path":"model-00004-of-00005.safetensors","componentId":"text_encoder","bytes":4931296656,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-it-qat-q4_0-unquantized?blobs=true"},{"repo":"google/gemma-3-12b-it-qat-q4_0-unquantized","path":"model-00005-of-00005.safetensors","componentId":"text_encoder","bytes":4601000928,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-it-qat-q4_0-unquantized?blobs=true"}],"sources":[{"label":"LTX-2.3 models list","url":"https://github.com/Lightricks/LTX-2/blob/main/MODELS-LTX-2.3.md","checkedAt":"2026-10-24","evidence":"Each checkpoint below is a single file bundling the"},{"label":"LTX-2.3 models list","url":"https://github.com/Lightricks/LTX-2/blob/main/MODELS-LTX-2.3.md","checkedAt":"2026-10-24","evidence":"https://huggingface.co/google/gemma-3-12b-it-qat-q4_0-unquantized/tree/main"}]},{"id":"fp8-runtime-4bit","label":"Official FP8 dev checkpoint, quantized to 4-bit at load","format":"safetensors","quantization":"4-bit at load from FP8","apps":["ltx-video-mlx"],"files":[{"repo":"Lightricks/LTX-2.3-fp8","path":"ltx-2.3-22b-dev-fp8.safetensors","componentId":"dit","bytes":29145431166,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Lightricks/LTX-2.3-fp8?blobs=true"},{"repo":"Lightricks/LTX-2.3","path":"ltx-2.3-22b-distilled-lora-384.safetensors","componentId":"distilled_lora","bytes":7605507256,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Lightricks/LTX-2.3?blobs=true"},{"repo":"google/gemma-3-12b-pt","path":"model-00001-of-00005.safetensors","componentId":"text_encoder","bytes":4979902192,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-pt?blobs=true"},{"repo":"google/gemma-3-12b-pt","path":"model-00002-of-00005.safetensors","componentId":"text_encoder","bytes":4931296592,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-pt?blobs=true"},{"repo":"google/gemma-3-12b-pt","path":"model-00003-of-00005.safetensors","componentId":"text_encoder","bytes":4931296656,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-pt?blobs=true"},{"repo":"google/gemma-3-12b-pt","path":"model-00004-of-00005.safetensors","componentId":"text_encoder","bytes":4931296656,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-pt?blobs=true"},{"repo":"google/gemma-3-12b-pt","path":"model-00005-of-00005.safetensors","componentId":"text_encoder","bytes":4601000928,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/google/gemma-3-12b-pt?blobs=true"}],"sources":[{"label":"appautomaton ltx-video-mlx README","url":"https://raw.githubusercontent.com/appautomaton/ltx-video-mlx/83b4881d70304013231ce03509227af9c0df500c/README.md","checkedAt":"2026-10-24","evidence":"LTX-2.3 FP8 (29 GB)"},{"label":"appautomaton ltx-video-mlx README","url":"https://raw.githubusercontent.com/appautomaton/ltx-video-mlx/83b4881d70304013231ce03509227af9c0df500c/README.md","checkedAt":"2026-10-24","evidence":"Gemma 3 12B (~24 GB)"}]},{"id":"gguf-q4-k-m","label":"Unsloth GGUF dev Q4_K_M with GGUF Gemma 3 (ComfyUI)","format":"gguf","quantization":"Q4_K_M","apps":["comfyui"],"files":[{"repo":"unsloth/LTX-2.3-GGUF","path":"ltx-2.3-22b-dev-Q4_K_M.gguf","componentId":"dit","bytes":14326856736,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/LTX-2.3-GGUF?blobs=true"},{"repo":"unsloth/LTX-2.3-GGUF","path":"vae/ltx-2.3-22b-dev_video_vae.safetensors","componentId":"video_vae","bytes":1452256522,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/LTX-2.3-GGUF?blobs=true"},{"repo":"unsloth/LTX-2.3-GGUF","path":"vae/ltx-2.3-22b-dev_audio_vae.safetensors","componentId":"audio_vae","bytes":364853140,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/LTX-2.3-GGUF?blobs=true"},{"repo":"unsloth/LTX-2.3-GGUF","path":"text_encoders/ltx-2.3-22b-dev_embeddings_connectors.safetensors","componentId":"text_connector","bytes":2312144712,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/LTX-2.3-GGUF?blobs=true"},{"repo":"unsloth/gemma-3-12b-it-qat-GGUF","path":"gemma-3-12b-it-qat-UD-Q4_K_XL.gguf","componentId":"text_encoder","bytes":7432229248,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/gemma-3-12b-it-qat-GGUF?blobs=true"},{"repo":"unsloth/gemma-3-12b-it-qat-GGUF","path":"mmproj-BF16.gguf","componentId":"text_encoder","bytes":854200448,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/gemma-3-12b-it-qat-GGUF?blobs=true"}],"sources":[{"label":"Unsloth LTX-2.3 GGUF card","url":"https://huggingface.co/unsloth/LTX-2.3-GGUF","checkedAt":"2026-10-24","evidence":"ltx-2.3-22b-dev-Q4_K_M.gguf"},{"label":"Unsloth LTX-2.3 GGUF card","url":"https://huggingface.co/unsloth/LTX-2.3-GGUF","checkedAt":"2026-10-24","evidence":"gemma-3-12b-it-qat-UD-Q4_K_XL.gguf"}]}],"publishedMeasurements":[{"id":"ltx-2-3-appautomaton-m5max-128gb-512x512-121f","modelId":"ltx-2-3","buildId":"fp8-runtime-4bit","appId":"ltx-video-mlx","machine":{"chip":"Apple M5 Max","ramGb":128},"workload":{"resolution":"512x512","frames":121,"durationSeconds":4.8,"steps":8,"quantization":"4-bit at load from FP8","notes":"T2V, end-to-end total with approximate figures (the source marks them with ~). The repository carries no run date, so the date shown is the pinned commit date."},"outcome":"success","timeSeconds":101,"timeText":"~101s","peakMemoryGb":53,"peakMemoryText":"~53 GB","publishedAt":"2026-04-11","grade":"A","source":{"label":"appautomaton ltx-video-mlx performance notes","url":"https://raw.githubusercontent.com/appautomaton/ltx-video-mlx/83b4881d70304013231ce03509227af9c0df500c/docs/performance.md","checkedAt":"2026-10-24","evidence":"| 121 | 4.8s | 4,096 | ~45s | ~8s | ~53 GB | ~101s |"}}],"quality":null,"sources":[{"label":"LTX-2.3 model card","url":"https://huggingface.co/Lightricks/LTX-2.3","checkedAt":"2026-10-24","evidence":"LTX-2.3 is a DiT-based audio-video foundation model designed to generate synchronized video and audio within a single model."},{"label":"LTX-2 Community License Agreement","url":"https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2","checkedAt":"2026-10-24","evidence":"License date: January 5, 2026"},{"label":"LTX-2 Community License Agreement","url":"https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2","checkedAt":"2026-10-24","evidence":"Entities with annual revenues of at least $10,000,000 (the"}]},{"id":"minimax-h3","name":"MiniMax H3","modality":"video","creator":"MiniMax","url":"https://modelfit.io/video/minimax-h3/","officialRepository":"https://huggingface.co/MiniMaxAI/MiniMax-H3","license":{"name":"MiniMax H3 Community License Agreement","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/LICENSE","commercial":"restricted","excludedTerritories":["European Union","United Kingdom","Republic of Korea","United States of America"],"note":"Separate prior written authorization is required when commercial products and services generate more than 20 million US dollars in yearly revenue."},"components":[{"id":"dit","name":"H3-Omni-Transformer","role":"dit","stage":"denoise","required":true,"parametersB":33,"sources":[{"label":"MiniMax-H3 model card","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3","checkedAt":"2026-10-24","evidence":"H3\\-Omni\\-Transformer is a 33B\\-parameter dense, single\\-stream Transformer, with approximately 13B parameters residing in AdaLN\\-related branches\\."}]},{"id":"text_encoder","name":"H3-Encoder (Qwen3-VL-32B)","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"MiniMax-H3 model card","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3","checkedAt":"2026-10-24","evidence":"The H3\\-Encoder uses the full pretrained weights of Qwen3\\-VL\\-32B and provides the hidden states from its 50th layer"}]},{"id":"video_vae","name":"H3-VisualVAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"MiniMax-H3 model card","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3","checkedAt":"2026-10-24","evidence":"spatial compression factor of 16×, a temporal compression factor of 4×, and 24 latent channels, denoted as f16t4d24"}]},{"id":"audio_vae","name":"H3-AudioVAE","role":"audio","stage":"decode","required":true,"sources":[{"label":"MiniMax-H3 model card","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3","checkedAt":"2026-10-24","evidence":"For each channel, H3-AudioVAE compresses 32 kHz audio into a sequence of latent tokens with a temporal rate of 40 Hz."}]},{"id":"turbo_lora","name":"Turbo LoRA","role":"other","stage":"denoise","required":false,"sources":[]}],"builds":[{"id":"official-bf16","label":"Official FL2VA BF16 (MiniMaxAI/MiniMax-H3)","format":"safetensors","quantization":"BF16","apps":["argus-mlx","h3c"],"files":[{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00001-of-00013.safetensors","componentId":"dit","bytes":5227812968,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00002-of-00013.safetensors","componentId":"dit","bytes":5164578856,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00003-of-00013.safetensors","componentId":"dit","bytes":5164578872,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00004-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00005-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00006-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00007-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00008-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00009-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00010-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00011-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00012-of-00013.safetensors","componentId":"dit","bytes":5164578896,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/transformer/model-00013-of-00013.safetensors","componentId":"dit","bytes":4242305176,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00001-of-00014.safetensors","componentId":"text_encoder","bytes":4932328944,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00002-of-00014.safetensors","componentId":"text_encoder","bytes":4875990528,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00003-of-00014.safetensors","componentId":"text_encoder","bytes":4875990552,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00004-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00005-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00006-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00007-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00008-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00009-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00010-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00011-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00012-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00013-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00014-of-00014.safetensors","componentId":"text_encoder","bytes":3270697008,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/video_vae/source/model.safetensors","componentId":"video_vae","bytes":10415548320,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/audio_vae/model.safetensors","componentId":"audio_vae","bytes":605429308,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"}],"sources":[{"label":"MiniMax-H3 model card","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3","checkedAt":"2026-10-24","evidence":"|MiniMax\\-H3 Base FL2VA|Text\\-to\\-Audio\\-Video \\(`t2va`\\), First/Last\\-Frame\\-to\\-Audio\\-Video \\(`fl2va`\\)|Text; optional first frame, last frame, or both|Video and audio|BF16|"},{"label":"Argus MLX route README","url":"https://raw.githubusercontent.com/Argus-AiTeam/minimax-h3-mac/7d2d00d81a51a38031adaac8bf4c2aabc307c6d2/README_EN.md","checkedAt":"2026-10-24","evidence":"With MLX weight streaming, this project runs the original MiniMax-H3 BF16 DiT and BF16 text encoder"}]},{"id":"argus-int8","label":"Argus calibrated INT8 DiT with the official text encoder and VAEs","format":"mlx","quantization":"INT8","apps":["argus-mlx"],"files":[{"repo":"water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","path":"model-00001-of-00008.safetensors","componentId":"dit","bytes":5345568851,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8?blobs=true"},{"repo":"water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","path":"model-00002-of-00008.safetensors","componentId":"dit","bytes":5349483016,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8?blobs=true"},{"repo":"water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","path":"model-00003-of-00008.safetensors","componentId":"dit","bytes":5272413031,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8?blobs=true"},{"repo":"water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","path":"model-00004-of-00008.safetensors","componentId":"dit","bytes":5358127869,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8?blobs=true"},{"repo":"water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","path":"model-00005-of-00008.safetensors","componentId":"dit","bytes":5349483010,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8?blobs=true"},{"repo":"water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","path":"model-00006-of-00008.safetensors","componentId":"dit","bytes":5272412252,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8?blobs=true"},{"repo":"water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","path":"model-00007-of-00008.safetensors","componentId":"dit","bytes":5336807607,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8?blobs=true"},{"repo":"water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","path":"model-00008-of-00008.safetensors","componentId":"dit","bytes":512733582,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00001-of-00014.safetensors","componentId":"text_encoder","bytes":4932328944,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00002-of-00014.safetensors","componentId":"text_encoder","bytes":4875990528,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00003-of-00014.safetensors","componentId":"text_encoder","bytes":4875990552,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00004-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00005-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00006-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00007-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00008-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00009-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00010-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00011-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00012-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00013-of-00014.safetensors","componentId":"text_encoder","bytes":4875990584,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/text_encoder/model-00014-of-00014.safetensors","componentId":"text_encoder","bytes":3270697008,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/video_vae/source/model.safetensors","componentId":"video_vae","bytes":10415548320,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"},{"repo":"MiniMaxAI/MiniMax-H3","path":"FL2VA/audio_vae/model.safetensors","componentId":"audio_vae","bytes":605429308,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/MiniMaxAI/MiniMax-H3?blobs=true"}],"sources":[{"label":"Argus calibrated INT8 DiT card","url":"https://huggingface.co/water1234/MiniMax-H3-MLX-Argus-Calibrated-INT8","checkedAt":"2026-10-24","evidence":"Weight-only affine INT8, group size 32"},{"label":"Argus MLX route README","url":"https://raw.githubusercontent.com/Argus-AiTeam/minimax-h3-mac/dffe3d664f2afd9439bd32157e926aa5494175a0/README_EN.md","checkedAt":"2026-10-24","evidence":"streamed layer by layer in MLX 4-bit"}]},{"id":"mlx-serve-4bit","label":"MLX Serve FL2VA 4-bit pack with Turbo LoRA","format":"mlx","quantization":"4-bit","apps":["mlx-serve"],"files":[{"repo":"ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit","path":"transformer.safetensors","componentId":"dit","bytes":18698813290,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit?blobs=true"},{"repo":"ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit","path":"text_encoder.safetensors","componentId":"text_encoder","bytes":15804791921,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit?blobs=true"},{"repo":"ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit","path":"video_vae.safetensors","componentId":"video_vae","bytes":5207808496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit?blobs=true"},{"repo":"ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit","path":"audio_vae.safetensors","componentId":"audio_vae","bytes":605254808,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit?blobs=true"},{"repo":"ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit","path":"turbo_lora.safetensors","componentId":"turbo_lora","bytes":779849816,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit?blobs=true"}],"sources":[{"label":"MLX Serve 4-bit pack card","url":"https://huggingface.co/ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit","checkedAt":"2026-10-24","evidence":"Self-contained: weights, both VAEs and the tokenizer in one directory."},{"label":"MLX Serve 4-bit pack card","url":"https://huggingface.co/ddalcu/MiniMax-H3-FL2VA-MLX-Serve-4bit","checkedAt":"2026-10-24","evidence":"4-bit affine (group size 64) conversion of MiniMax-H3's FL2VA checkpoint for"}]},{"id":"gguf-q4-k","label":"Unsloth GGUF Q4_K denoiser with the Q4_K_M Qwen3-VL encoder","format":"gguf","quantization":"Q4_K","apps":["stable-diffusion-cpp"],"files":[{"repo":"unsloth/MiniMax-H3-GGUF","path":"minimax_h3_fl2va_pruned-Q4_K.gguf","componentId":"dit","bytes":11420663904,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/MiniMax-H3-GGUF?blobs=true"},{"repo":"unsloth/MiniMax-H3-GGUF","path":"qwen3vl_32b_minimax_h3-Q4_K_M.gguf","componentId":"text_encoder","bytes":18218065024,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/MiniMax-H3-GGUF?blobs=true"},{"repo":"Comfy-Org/MiniMax-H3","path":"vae/minimax_h3_video_vae_fp16.safetensors","componentId":"video_vae","bytes":5207808496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/MiniMax-H3?blobs=true"},{"repo":"Comfy-Org/MiniMax-H3","path":"vae/minimax_h3_audio_vae_fp32.safetensors","componentId":"audio_vae","bytes":605254808,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/MiniMax-H3?blobs=true"}],"sources":[{"label":"Unsloth MiniMax-H3 GGUF card","url":"https://huggingface.co/unsloth/MiniMax-H3-GGUF","checkedAt":"2026-10-24","evidence":"sd-cli --mode vid_gen"},{"label":"Unsloth MiniMax-H3 GGUF card","url":"https://huggingface.co/unsloth/MiniMax-H3-GGUF","checkedAt":"2026-10-24","evidence":"The VAEs are not duplicated here, take them from"}]},{"id":"comfyui-gguf-q5-k-m","label":"ComfyUI GGUF Q5_K_M pruned denoiser with Parasyte Turbo (Bambushu workflow)","format":"gguf","quantization":"Q5_K_M","apps":["comfyui"],"files":[{"repo":"Abiray/MiniMax-H3-Pruned-GGUF","path":"MiniMax-H3-FL2VA-Pruned-Q5_K_M.gguf","componentId":"dit","bytes":14068966496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Abiray/MiniMax-H3-Pruned-GGUF?blobs=true"},{"repo":"realrebelai/MiniMax-H3_GGUFs","path":"qwen3vl-32B-MiniMax-H3-Q4_K_M.gguf","componentId":"text_encoder","bytes":14576977888,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/realrebelai/MiniMax-H3_GGUFs?blobs=true"},{"repo":"Comfy-Org/MiniMax-H3","path":"vae/minimax_h3_video_vae_fp16.safetensors","componentId":"video_vae","bytes":5207808496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/MiniMax-H3?blobs=true"},{"repo":"Comfy-Org/MiniMax-H3","path":"vae/minimax_h3_audio_vae_fp32.safetensors","componentId":"audio_vae","bytes":605254808,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/MiniMax-H3?blobs=true"},{"repo":"Plaguekind/H3-Lora","path":"H3-PK-Parasyte-Turbo.safetensors","componentId":"turbo_lora","bytes":2115774558,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Plaguekind/H3-Lora?blobs=true"}],"sources":[{"label":"Bambushu minimax-h3-mac README","url":"https://github.com/Bambushu/minimax-h3-mac/blob/6bc2f75efbb8ef6c79b70660c304f1b930a04398/README.md","checkedAt":"2026-10-24","evidence":"MiniMax-H3-FL2VA-Pruned-Q5_K_M.gguf"},{"label":"Bambushu minimax-h3-mac README","url":"https://github.com/Bambushu/minimax-h3-mac/blob/6bc2f75efbb8ef6c79b70660c304f1b930a04398/README.md","checkedAt":"2026-10-24","evidence":"qwen3vl-32B-MiniMax-H3-Q4_K_M.gguf"},{"label":"Bambushu minimax-h3-mac README","url":"https://github.com/Bambushu/minimax-h3-mac/blob/6bc2f75efbb8ef6c79b70660c304f1b930a04398/README.md","checkedAt":"2026-10-24","evidence":"H3-PK-Parasyte-Turbo.safetensors"}]},{"id":"comfyui-int8-pruned","label":"ComfyUI pruned INT8 convrot denoiser with the NVFP4 encoder (rznotes setup)","format":"safetensors","quantization":"INT8 convrot","apps":["comfyui"],"files":[{"repo":"Comfy-Org/MiniMax-H3","path":"diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors","componentId":"dit","bytes":20970379616,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/MiniMax-H3?blobs=true"},{"repo":"Comfy-Org/MiniMax-H3","path":"text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors","componentId":"text_encoder","bytes":15687142551,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/MiniMax-H3?blobs=true"},{"repo":"Comfy-Org/MiniMax-H3","path":"vae/minimax_h3_video_vae_fp16.safetensors","componentId":"video_vae","bytes":5207808496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/MiniMax-H3?blobs=true"},{"repo":"Comfy-Org/MiniMax-H3","path":"vae/minimax_h3_audio_vae_fp32.safetensors","componentId":"audio_vae","bytes":605254808,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/MiniMax-H3?blobs=true"}],"sources":[{"label":"rznotes: MiniMax H3 with ComfyUI on an M5 Pro 64GB","url":"https://rznotes.com/blog/minimax-h3-m5-pro-comfyui/","checkedAt":"2026-10-24","evidence":"minimax_h3_fl2va_pruned_int8_convrot.safetensors"},{"label":"rznotes: MiniMax H3 with ComfyUI on an M5 Pro 64GB","url":"https://rznotes.com/blog/minimax-h3-m5-pro-comfyui/","checkedAt":"2026-10-24","evidence":"qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors"}]}],"publishedMeasurements":[{"id":"minimax-h3-comfyui-light-m1air-16gb-352x608-22f-4steps","modelId":"minimax-h3","appId":"comfyui","machine":{"device":"MacBook Air","chip":"Apple M1","ramGb":16},"workload":{"resolution":"352x608","frames":22,"steps":4,"notes":"H3 Turbo LoRA merged in, 4 steps, low_vram on. Light setup: Q3_K_M GGUF denoiser, ClipProj 4B encoder on the CPU, audio enabled. The date is the README commit date; the run date is not stated."},"outcome":"success","timeSeconds":1020,"timeText":"17 min","publishedAt":"2026-10-22","grade":"B","source":{"label":"Bambushu minimax-h3-mac README","url":"https://github.com/Bambushu/minimax-h3-mac/blob/6bc2f75efbb8ef6c79b70660c304f1b930a04398/README.md","checkedAt":"2026-10-24","evidence":"17 min with the [H3 Turbo LoRA](https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo) merged in (4 steps, `low_vram` on)"}},{"id":"minimax-h3-comfyui-light-m1air-16gb-352x608-22f-8steps","modelId":"minimax-h3","appId":"comfyui","machine":{"device":"MacBook Air","chip":"Apple M1","ramGb":16},"workload":{"resolution":"352x608","frames":22,"steps":8,"notes":"Parasyte LoRA, 8 steps. Light setup: Q3_K_M GGUF denoiser, ClipProj 4B encoder on the CPU, audio enabled. The date is the README commit date; the run date is not stated."},"outcome":"success","timeSeconds":2160,"timeText":"36 min","publishedAt":"2026-10-22","grade":"B","source":{"label":"Bambushu minimax-h3-mac README","url":"https://github.com/Bambushu/minimax-h3-mac/blob/6bc2f75efbb8ef6c79b70660c304f1b930a04398/README.md","checkedAt":"2026-10-24","evidence":"22 frames (~1s) with audio in 36 min with Parasyte (8 steps)"}},{"id":"minimax-h3-comfyui-light-m1air-16gb-352x608-56f-failed","modelId":"minimax-h3","appId":"comfyui","machine":{"device":"MacBook Air","chip":"Apple M1","ramGb":16},"workload":{"resolution":"352x608","frames":56,"notes":"56 frames and more. Light setup: Q3_K_M GGUF denoiser, ClipProj 4B encoder on the CPU, audio enabled. The date is the README commit date; the run date is not stated."},"outcome":"failed","publishedAt":"2026-10-22","grade":"B","source":{"label":"Bambushu minimax-h3-mac README","url":"https://github.com/Bambushu/minimax-h3-mac/blob/6bc2f75efbb8ef6c79b70660c304f1b930a04398/README.md","checkedAt":"2026-10-24","evidence":"56 frames and up ran out of GPU memory."}},{"id":"minimax-h3-comfyui-gguf-q5-k-m-m5pro-48gb-352x608-124f","modelId":"minimax-h3","buildId":"comfyui-gguf-q5-k-m","appId":"comfyui","machine":{"chip":"Apple M5 Pro","ramGb":48},"workload":{"resolution":"352x608","frames":124,"fps":24,"durationSeconds":5.17,"steps":8,"quantization":"Q5_K_M","notes":"Pruned FL2VA GGUF Q5_K_M with Qwen3-VL 32B Q4_K_M, Parasyte Turbo strength 1.5, ER-SDE beta57, sigma shift video 12 and audio 3, audio enabled, seed 77001. Time includes model loading, sampling, decoding and saving; one run in an existing session. PyTorch 2.13.0, ComfyUI v0.34.0 plus 11 commits."},"outcome":"success","timeSeconds":414,"timeText":"6m 54s","publishedAt":"2026-09-14","grade":"A","source":{"label":"Bambushu minimax-h3-mac README","url":"https://github.com/Bambushu/minimax-h3-mac/blob/6bc2f75efbb8ef6c79b70660c304f1b930a04398/README.md","checkedAt":"2026-10-24","evidence":"124 frames / 5.17s | **6m 54s** |"}},{"id":"minimax-h3-comfyui-gguf-q5-k-m-m5pro-48gb-480x864-124f","modelId":"minimax-h3","buildId":"comfyui-gguf-q5-k-m","appId":"comfyui","machine":{"chip":"Apple M5 Pro","ramGb":48},"workload":{"resolution":"480x864","frames":124,"fps":24,"durationSeconds":5.17,"steps":8,"quantization":"Q5_K_M","notes":"Pruned FL2VA GGUF Q5_K_M with Qwen3-VL 32B Q4_K_M, Parasyte Turbo strength 1.5, ER-SDE beta57, sigma shift video 12 and audio 3, audio enabled, seed 77001. Time includes model loading, sampling, decoding and saving; one run in an existing session. PyTorch 2.13.0, ComfyUI v0.34.0 plus 11 commits."},"outcome":"success","timeSeconds":772,"timeText":"12m 52s","publishedAt":"2026-09-14","grade":"A","source":{"label":"Bambushu minimax-h3-mac README","url":"https://github.com/Bambushu/minimax-h3-mac/blob/6bc2f75efbb8ef6c79b70660c304f1b930a04398/README.md","checkedAt":"2026-10-24","evidence":"| 0.4 MP | 480×864 | 124 frames / 5.17s | **12m 52s** |"}},{"id":"minimax-h3-comfyui-int8-pruned-m5pro-64gb-352x608-124f","modelId":"minimax-h3","buildId":"comfyui-int8-pruned","appId":"comfyui","machine":{"device":"MacBook Pro","chip":"Apple M5 Pro","ramGb":64},"workload":{"resolution":"352x608","frames":124,"durationSeconds":5.167,"steps":10,"quantization":"INT8 convrot","notes":"First and last frame image-to-video with the model audio track. ComfyUI v0.31.0 with the ComfyUI-AppleSilicon-FP8 custom node, macOS 26.5. The date is the article date; the run date is not stated."},"outcome":"success","timeSeconds":501.6,"publishedAt":"2026-08-13","grade":"A","source":{"label":"rznotes: MiniMax H3 with ComfyUI on an M5 Pro 64GB","url":"https://rznotes.com/blog/minimax-h3-m5-pro-comfyui/","checkedAt":"2026-10-24","evidence":"501.6 秒（8 分 22 秒）"}},{"id":"minimax-h3-h3c-bf16-m5pro-64gb-576x1024-124f","modelId":"minimax-h3","buildId":"official-bf16","appId":"h3c","machine":{"device":"MacBook Pro","chip":"Apple M5 Pro","ramGb":64},"workload":{"resolution":"576x1024","frames":124,"durationSeconds":5.167,"quantization":"BF16","notes":"Total generation time with the full BF16 checkpoint; the step count is not stated. h3.c commit 03cb133, macOS 26.6.1. The date is the article date; the run date is not stated."},"outcome":"success","timeSeconds":1859.12,"publishedAt":"2026-08-13","grade":"A","source":{"label":"rznotes: h3.c on an M5 Pro 64GB","url":"https://rznotes.com/blog/h3c-minimax-h3-m5-pro/","checkedAt":"2026-10-24","evidence":"1859.12 秒（約 30 分 59 秒）"}},{"id":"minimax-h3-h3c-int8-m5pro-64gb-576x1024-124f","modelId":"minimax-h3","appId":"h3c","machine":{"device":"MacBook Pro","chip":"Apple M5 Pro","ramGb":64},"workload":{"resolution":"576x1024","frames":124,"durationSeconds":5.167,"quantization":"int8 at load (h3.c)","notes":"Same checkpoint as the BF16 run with the h3.c int8 path, so it is not a separate download and is not tied to a build. Total generation time; the step count is not stated. h3.c commit 03cb133, macOS 26.6.1. The date is the article date."},"outcome":"success","timeSeconds":1439.96,"publishedAt":"2026-08-13","grade":"B","source":{"label":"rznotes: h3.c on an M5 Pro 64GB","url":"https://rznotes.com/blog/h3c-minimax-h3-m5-pro/","checkedAt":"2026-10-24","evidence":"1439.96 秒（約 23 分 59 秒）"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-384x224-56f","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"384x224","frames":56,"durationSeconds":2.3,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA, audio enabled, same prompt and seed. Generation time as reported; the run date is not stated."},"outcome":"success","timeSeconds":31.2,"peakMemoryGb":66.7,"publishedAt":"2026-08-12","grade":"A","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"<td>31.2s</td>"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-512x288-56f","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"512x288","frames":56,"durationSeconds":2.3,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA, audio enabled, same prompt and seed. Generation time as reported; the run date is not stated."},"outcome":"success","timeSeconds":53.8,"peakMemoryGb":68.6,"publishedAt":"2026-08-12","grade":"A","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"<td>53.8s</td>"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-608x352-124f","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"608x352","frames":124,"durationSeconds":5.2,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA, audio enabled, same prompt and seed. Generation time as reported; the run date is not stated."},"outcome":"success","timeSeconds":310.1,"peakMemoryGb":70.4,"publishedAt":"2026-08-12","grade":"A","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"<td>310.1s</td>"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-608x352-226f","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"608x352","frames":226,"durationSeconds":9.4,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA, audio enabled, same prompt and seed. Generation time as reported; the run date is not stated."},"outcome":"success","timeSeconds":654,"peakMemoryGb":76.6,"publishedAt":"2026-08-12","grade":"A","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"<td>654.0s</td>"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-608x352-22f","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"608x352","frames":22,"durationSeconds":0.9,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA, audio enabled, same prompt and seed. Generation time as reported; the run date is not stated."},"outcome":"success","timeSeconds":33.3,"peakMemoryGb":66.5,"publishedAt":"2026-08-12","grade":"A","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"<td>33.3s</td>"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-608x352-362f-failed","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"608x352","frames":362,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA. The run swapped under memory pressure and the first diffusion step alone took 79 minutes."},"outcome":"failed","publishedAt":"2026-08-12","grade":"B","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"メモリが逼迫してスワップし、拡散の 1 ステップ目だけで 79 分。"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-608x352-56f","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"608x352","frames":56,"durationSeconds":2.3,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA, audio enabled, same prompt and seed. Generation time as reported; the run date is not stated."},"outcome":"success","timeSeconds":75.2,"peakMemoryGb":67.6,"publishedAt":"2026-08-12","grade":"A","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"<td>75.2s</td>"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-768x448-56f","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"768x448","frames":56,"durationSeconds":2.3,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA, audio enabled, same prompt and seed. Generation time as reported; the run date is not stated."},"outcome":"success","timeSeconds":114.3,"peakMemoryGb":72.8,"publishedAt":"2026-08-12","grade":"A","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"<td>114.3s</td>"}},{"id":"minimax-h3-mlx-serve-4bit-m4max-128gb-960x544-56f","modelId":"minimax-h3","buildId":"mlx-serve-4bit","appId":"mlx-serve","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":128},"workload":{"resolution":"960x544","frames":56,"durationSeconds":2.3,"steps":4,"quantization":"4-bit","notes":"Turbo LoRA, audio enabled, same prompt and seed. Generation time as reported; the run date is not stated."},"outcome":"success","timeSeconds":203.3,"peakMemoryGb":75.6,"publishedAt":"2026-08-12","grade":"A","source":{"label":"Zenn: MiniMax H3 on M4 Max 128GB","url":"https://zenn.dev/acntechjp/articles/zenn-minimax-h3-m4max","checkedAt":"2026-10-24","evidence":"<td>203.3s</td>"}},{"id":"minimax-h3-argus-bf16-m4pro-24gb","modelId":"minimax-h3","buildId":"official-bf16","appId":"argus-mlx","machine":{"device":"MacBook Pro (Mac16,8)","chip":"Apple M4 Pro","ramGb":24},"workload":{"resolution":"1344x768","frames":124,"fps":24,"durationSeconds":5.17,"steps":5,"quantization":"BF16","notes":"Five sigma points are four DiT forwards. LightX2V 4-step 768p Turbo adapter, two streamed blocks, no block cache, audio enabled. Peak memory footprint is approximate. The date is the source revision date; the run date is not stated."},"outcome":"success","timeSeconds":2878.7,"timeText":"2,878.7 s","peakMemoryGb":15.8,"publishedAt":"2026-08-11","grade":"A","source":{"label":"Argus MLX route README (BF16 run)","url":"https://raw.githubusercontent.com/Argus-AiTeam/minimax-h3-mac/7d2d00d81a51a38031adaac8bf4c2aabc307c6d2/README_EN.md","checkedAt":"2026-10-24","evidence":"| **End-to-end** | **2,878.7 s (47 min 58.7 s)** |"}},{"id":"minimax-h3-argus-int8-m4pro-24gb","modelId":"minimax-h3","buildId":"argus-int8","appId":"argus-mlx","machine":{"device":"MacBook Pro (Mac16,8)","chip":"Apple M4 Pro","ramGb":24},"workload":{"resolution":"768x448","frames":124,"fps":24,"durationSeconds":5.1667,"steps":9,"quantization":"INT8","notes":"Nine sigma points are eight DiT forwards. Calibrated INT8 DiT, text encoder streamed in 4-bit, native BF16 MLX Turbo, streaming and no block cache. Peak is maximum RSS, approximate. The date is the source revision date; the run date is not stated."},"outcome":"success","timeSeconds":1378,"timeText":"1,378.0 s","peakMemoryGb":12,"peakMemoryText":"12.0 GB","publishedAt":"2026-08-11","grade":"A","source":{"label":"Argus MLX route README (INT8 run)","url":"https://raw.githubusercontent.com/Argus-AiTeam/minimax-h3-mac/dffe3d664f2afd9439bd32157e926aa5494175a0/README_EN.md","checkedAt":"2026-10-24","evidence":"| **End-to-end** | **1,378.0 s (approximately 23 minutes)** |"}}],"quality":{"modelId":"minimax-h3","category":"text_to_video","sourceName":"minimax-h3","rank":8,"total":48,"votes":11417,"rating":1460.3432579513055,"low":1451.7376004067319,"high":1468.9489154958787,"snapshot":"2026-09-22","attribution":"Blind human votes from Arena (https://arena.ai), CC BY 4.0. Arena checkpoint, not a quantized local-build score.","dataset":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset"},"sources":[{"label":"MiniMax-H3 model card","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3","checkedAt":"2026-10-24","evidence":"|MiniMax\\-H3 Base FL2VA|Text\\-to\\-Audio\\-Video \\(`t2va`\\), First/Last\\-Frame\\-to\\-Audio\\-Video \\(`fl2va`\\)|Text; optional first frame, last frame, or both|Video and audio|BF16|"},{"label":"MiniMax H3 Community License Agreement","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/LICENSE","checkedAt":"2026-10-24","evidence":"5. “Excluded Territories” means the European Union, the United Kingdom, the Republic of Korea and the United States of America."},{"label":"MiniMax H3 Community License Agreement","url":"https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/LICENSE","checkedAt":"2026-10-24","evidence":"if your commercial products and services generate more than 20 million US dollars (or equivalent in other currencies) in yearly revenue."}]},{"id":"mochi-1","name":"Mochi 1 preview","modality":"video","creator":"Genmo","url":"https://modelfit.io/video/mochi-1/","officialRepository":"https://huggingface.co/genmo/mochi-1-preview","license":{"name":"Apache License 2.0","url":"https://huggingface.co/genmo/mochi-1-preview/blob/main/LICENSE","commercial":"yes","excludedTerritories":[]},"components":[{"id":"dit","name":"AsymmDiT","role":"dit","stage":"denoise","required":true,"parametersB":10,"sources":[{"label":"Mochi 1 model card","url":"https://huggingface.co/genmo/mochi-1-preview","checkedAt":"2026-10-24","evidence":"featuring a 10 billion parameter diffusion model built on our novel Asymmetric Diffusion Transformer (AsymmDiT) architecture"}]},{"id":"text_encoder","name":"T5-XXL text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"Mochi 1 model card","url":"https://huggingface.co/genmo/mochi-1-preview","checkedAt":"2026-10-24","evidence":"Mochi 1 simply encodes prompts with a single T5-XXL language model."}]},{"id":"video_vae","name":"AsymmVAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"Mochi 1 model card","url":"https://huggingface.co/genmo/mochi-1-preview","checkedAt":"2026-10-24","evidence":"|362M   | 64  | 128  | 12   | 8x8   | 6x   |"}]}],"builds":[{"id":"diffusers-bf16","label":"Official Diffusers BF16 variant","format":"safetensors","quantization":"BF16","apps":["diffusers"],"files":[{"repo":"genmo/mochi-1-preview","path":"transformer/diffusion_pytorch_model.bf16-00001-of-00003.safetensors","componentId":"dit","bytes":9937333040,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/genmo/mochi-1-preview?blobs=true"},{"repo":"genmo/mochi-1-preview","path":"transformer/diffusion_pytorch_model.bf16-00002-of-00003.safetensors","componentId":"dit","bytes":9929099008,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/genmo/mochi-1-preview?blobs=true"},{"repo":"genmo/mochi-1-preview","path":"transformer/diffusion_pytorch_model.bf16-00003-of-00003.safetensors","componentId":"dit","bytes":189051592,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/genmo/mochi-1-preview?blobs=true"},{"repo":"genmo/mochi-1-preview","path":"text_encoder/model-00001-of-00004.safetensors","componentId":"text_encoder","bytes":4989319680,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/genmo/mochi-1-preview?blobs=true"},{"repo":"genmo/mochi-1-preview","path":"text_encoder/model-00002-of-00004.safetensors","componentId":"text_encoder","bytes":4999830656,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/genmo/mochi-1-preview?blobs=true"},{"repo":"genmo/mochi-1-preview","path":"text_encoder/model-00003-of-00004.safetensors","componentId":"text_encoder","bytes":4865612720,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/genmo/mochi-1-preview?blobs=true"},{"repo":"genmo/mochi-1-preview","path":"text_encoder/model-00004-of-00004.safetensors","componentId":"text_encoder","bytes":4194506688,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/genmo/mochi-1-preview?blobs=true"},{"repo":"genmo/mochi-1-preview","path":"vae/diffusion_pytorch_model.bf16.safetensors","componentId":"video_vae","bytes":919549966,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/genmo/mochi-1-preview?blobs=true"}],"sources":[{"label":"Mochi 1 model card","url":"https://huggingface.co/genmo/mochi-1-preview","checkedAt":"2026-10-24","evidence":"requires 22GB VRAM to run. There is a slight drop in the quality of the generated video as a result."}]},{"id":"comfyui-bf16","label":"Comfy-Org repackaged BF16 with FP16 T5","format":"safetensors","quantization":"BF16","apps":["comfyui"],"files":[{"repo":"Comfy-Org/Mochi-Preview","path":"split_files/diffusion_models/mochi_preview_bf16.safetensors","componentId":"dit","bytes":20055467768,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Mochi-Preview?blobs=true"},{"repo":"Comfy-Org/Mochi-Preview","path":"split_files/text_encoders/t5xxl_fp16.safetensors","componentId":"text_encoder","bytes":9787841024,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Mochi-Preview?blobs=true"},{"repo":"Comfy-Org/Mochi-Preview","path":"split_files/vae/mochi_vae.safetensors","componentId":"video_vae","bytes":919544702,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Mochi-Preview?blobs=true"}],"sources":[{"label":"Comfy-Org Mochi Preview repackaged card","url":"https://huggingface.co/Comfy-Org/Mochi-Preview","checkedAt":"2026-10-24","evidence":"mochi_preview_bf16.safetensors"}]},{"id":"comfyui-fp8-scaled","label":"Comfy-Org repackaged FP8 scaled","format":"safetensors","quantization":"FP8 scaled","apps":["comfyui"],"files":[{"repo":"Comfy-Org/Mochi-Preview","path":"split_files/diffusion_models/mochi_preview_fp8_scaled.safetensors","componentId":"dit","bytes":10028084160,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Mochi-Preview?blobs=true"},{"repo":"Comfy-Org/Mochi-Preview","path":"split_files/text_encoders/t5xxl_fp8_e4m3fn_scaled.safetensors","componentId":"text_encoder","bytes":5157348688,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Mochi-Preview?blobs=true"},{"repo":"Comfy-Org/Mochi-Preview","path":"split_files/vae/mochi_vae.safetensors","componentId":"video_vae","bytes":919544702,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Mochi-Preview?blobs=true"}],"sources":[{"label":"Comfy-Org Mochi Preview repackaged card","url":"https://huggingface.co/Comfy-Org/Mochi-Preview","checkedAt":"2026-10-24","evidence":"mochi_preview_fp8_scaled.safetensors"}]},{"id":"gguf-q3-k-m","label":"calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE","format":"gguf","quantization":"Q3_K_M","apps":["comfyui"],"files":[{"repo":"calcuis/mochi","path":"mochi-q3_k_m.gguf","componentId":"dit","bytes":4310877600,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/calcuis/mochi?blobs=true"},{"repo":"calcuis/mochi","path":"t5xxl_fp16-q4_0.gguf","componentId":"text_encoder","bytes":2896123072,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/calcuis/mochi?blobs=true"},{"repo":"calcuis/mochi","path":"mochi_vae_fp8_e4m3fn.safetensors","componentId":"video_vae","bytes":459794907,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/calcuis/mochi?blobs=true"}],"sources":[{"label":"calcuis mochi GGUF card","url":"https://huggingface.co/calcuis/mochi","checkedAt":"2026-10-24","evidence":"drag mochi-q3_k_m.gguf"},{"label":"calcuis mochi GGUF card","url":"https://huggingface.co/calcuis/mochi","checkedAt":"2026-10-24","evidence":"drag t5xxl_fp16-q4_0.gguf"}]}],"publishedMeasurements":[{"id":"mochi-1-diffusers-bf16-m4max-64gb","modelId":"mochi-1","buildId":"diffusers-bf16","appId":"diffusers","machine":{"device":"MacBook Pro","chip":"Apple M4 Max","ramGb":64},"workload":{"resolution":"848x480","frames":84,"durationSeconds":2.8,"steps":64,"quantization":"BF16","notes":"Diffusers MochiPipeline bf16 variant on PyTorch MPS, seed 42, one run. Swap of 19.8 GB reported. Peak is sampled every 0.5 s. Sampler, CFG and attention implementation are not reported. The date is the article date; the run date is not stated."},"outcome":"success","timeSeconds":8940.3,"peakMemoryGb":94.5,"publishedAt":"2026-09-02","grade":"A","source":{"label":"Zenn: Mochi 1 on M4 Max 64GB","url":"https://zenn.dev/taka4/articles/1ff69091c5dbb6","checkedAt":"2026-10-24","evidence":"8940.3"}}],"quality":{"modelId":"mochi-1","category":"text_to_video","sourceName":"mochi-v1","rank":48,"total":48,"votes":5893,"rating":1006.6195115514162,"low":989.9645901762071,"high":1023.2744329266251,"snapshot":"2026-09-22","attribution":"Blind human votes from Arena (https://arena.ai), CC BY 4.0. Arena checkpoint, not a quantized local-build score.","dataset":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset"},"sources":[{"label":"Mochi 1 model card","url":"https://huggingface.co/genmo/mochi-1-preview","checkedAt":"2026-10-24","evidence":"The initial release generates videos at 480p today."},{"label":"Apache License 2.0","url":"https://huggingface.co/genmo/mochi-1-preview/blob/main/LICENSE","checkedAt":"2026-10-24","evidence":"Version 2.0, January 2004"}]},{"id":"qwen-image-2-1","name":"Qwen-Image-2.1","modality":"image","creator":"Qwen","url":"https://modelfit.io/image/qwen-image-2-1/","officialRepository":"https://huggingface.co/Qwen/Qwen-Image-2.1","license":{"name":"Qwen Research License Agreement","url":"https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE","commercial":"no","excludedTerritories":[],"note":"Commercial use needs a separate commercial license from Qwen."},"components":[{"id":"dit","name":"Qwen-Image-2.1 transformer","role":"dit","stage":"denoise","required":true,"sources":[{"label":"Qwen-Image-2.1 pipeline config","url":"https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/model_index.json","checkedAt":"2026-10-24","evidence":"QwenImage21Transformer2DModel"}],"parametersB":7},{"id":"text_encoder","name":"Qwen3-VL text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"Qwen-Image-2.1 pipeline config","url":"https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/model_index.json","checkedAt":"2026-10-24","evidence":"Qwen3VLForConditionalGeneration"}]},{"id":"vae","name":"VAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"Qwen-Image-2.1 pipeline config","url":"https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/model_index.json","checkedAt":"2026-10-24","evidence":"AutoencoderKLQwenImage21"}]}],"builds":[{"id":"bf16-diffusers","label":"BF16 safetensors","format":"safetensors","quantization":"BF16","apps":["diffusers-staged","diffusers"],"files":[{"repo":"Qwen/Qwen-Image-2.1","path":"transformer/diffusion_pytorch_model-00001-of-00002.safetensors","componentId":"dit","bytes":9968332504,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"},{"repo":"Qwen/Qwen-Image-2.1","path":"transformer/diffusion_pytorch_model-00002-of-00002.safetensors","componentId":"dit","bytes":4261951904,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"},{"repo":"Qwen/Qwen-Image-2.1","path":"text_encoder/model-00001-of-00004.safetensors","componentId":"text_encoder","bytes":4998056552,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"},{"repo":"Qwen/Qwen-Image-2.1","path":"text_encoder/model-00002-of-00004.safetensors","componentId":"text_encoder","bytes":4915962464,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"},{"repo":"Qwen/Qwen-Image-2.1","path":"text_encoder/model-00003-of-00004.safetensors","componentId":"text_encoder","bytes":4915962496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"},{"repo":"Qwen/Qwen-Image-2.1","path":"text_encoder/model-00004-of-00004.safetensors","componentId":"text_encoder","bytes":2704357976,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"},{"repo":"Qwen/Qwen-Image-2.1","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":1350989512,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"}],"sources":[{"label":"Qwen-Image-2.1 model card","url":"https://huggingface.co/Qwen/Qwen-Image-2.1","checkedAt":"2026-10-24","evidence":"from diffusers import QwenImage21Pipeline"},{"label":"How to run Qwen-Image-2.1 on a 48 GB Mac without swap","url":"https://dev.to/nefayran/how-to-run-qwen-image-21-on-a-48-gb-mac-without-swap-3g4b","checkedAt":"2026-10-24","evidence":"Every run made one 1024 × 1024 image in 20 steps with seed 7"}]},{"id":"gguf-q8_0","label":"Q8_0 GGUF","format":"gguf","quantization":"Q8_0","apps":["stable-diffusion-cpp"],"files":[{"repo":"unsloth/Qwen-Image-2.1-GGUF","path":"qwen-image-2.1-Q8_0.gguf","componentId":"dit","bytes":7640860384,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/Qwen-Image-2.1-GGUF?blobs=true"},{"repo":"unsloth/Qwen3-VL-8B-Instruct-GGUF","path":"Qwen3-VL-8B-Instruct-UD-Q4_K_XL.gguf","componentId":"text_encoder","bytes":5148699488,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/Qwen3-VL-8B-Instruct-GGUF?blobs=true"},{"repo":"Qwen/Qwen-Image-2.1","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":1350989512,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"}],"sources":[{"label":"Qwen-Image-2.1 GGUF card","url":"https://huggingface.co/unsloth/Qwen-Image-2.1-GGUF","checkedAt":"2026-10-24","evidence":"Run these with [Unsloth Desktop](https://github.com/unslothai/unsloth), stable-diffusion.cpp and more. A GGUF is the denoiser only, so it needs the VAE and the Qwen3-VL text encoder alongside it."},{"label":"stable-diffusion.cpp Qwen Image 2.1 guide","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/docs/qwen_image_2.1.md","checkedAt":"2026-10-24","evidence":"Qwen Image 2.1 supports text-to-image generation and image editing, using Qwen3-VL-8B as the text encoder and its own VAE."}]},{"id":"gguf-q4_k_m","label":"Q4_K_M GGUF","format":"gguf","quantization":"Q4_K_M","apps":["stable-diffusion-cpp"],"files":[{"repo":"unsloth/Qwen-Image-2.1-GGUF","path":"qwen-image-2.1-Q4_K_M.gguf","componentId":"dit","bytes":4199565024,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/Qwen-Image-2.1-GGUF?blobs=true"},{"repo":"unsloth/Qwen3-VL-8B-Instruct-GGUF","path":"Qwen3-VL-8B-Instruct-UD-Q4_K_XL.gguf","componentId":"text_encoder","bytes":5148699488,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/Qwen3-VL-8B-Instruct-GGUF?blobs=true"},{"repo":"Qwen/Qwen-Image-2.1","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":1350989512,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Qwen/Qwen-Image-2.1?blobs=true"}],"sources":[{"label":"Qwen-Image-2.1 GGUF card","url":"https://huggingface.co/unsloth/Qwen-Image-2.1-GGUF","checkedAt":"2026-10-24","evidence":"Run these with [Unsloth Desktop](https://github.com/unslothai/unsloth), stable-diffusion.cpp and more. A GGUF is the denoiser only, so it needs the VAE and the Qwen3-VL text encoder alongside it."},{"label":"stable-diffusion.cpp Qwen Image 2.1 guide","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/docs/qwen_image_2.1.md","checkedAt":"2026-10-24","evidence":"Qwen Image 2.1 supports text-to-image generation and image editing, using Qwen3-VL-8B as the text encoder and its own VAE."}]}],"publishedMeasurements":[{"id":"qwen-image-2-1-diffusers-staged-m5-pro-48gb-1024","modelId":"qwen-image-2-1","buildId":"bf16-diffusers","appId":"diffusers-staged","machine":{"chip":"Apple M5 Pro","ramGb":48},"workload":{"resolution":"1024x1024","steps":20,"quantization":"BF16","notes":"Seed 7, torch 2.14.0, diffusers 0.41.0.dev0, one model in memory per stage. Two runs finished in 94 s and 124 s; the slower is shown. The footprint peaked at 19.0 GB while encoding."},"outcome":"success","timeSeconds":124,"peakMemoryGb":19,"peakMemoryText":"19.0 GB","publishedAt":"2026-10-11","grade":"A","source":{"label":"How to run Qwen-Image-2.1 on a 48 GB Mac without swap","url":"https://dev.to/nefayran/how-to-run-qwen-image-21-on-a-48-gb-mac-without-swap-3g4b","checkedAt":"2026-10-24","evidence":"The runs used an Apple M5 Pro with 48 GB, torch 2.14.0 and diffusers 0.41.0.dev0 from main."}}],"quality":{"modelId":"qwen-image-2-1","category":"text_to_image","sourceName":"qwen-image-2.1","rank":19,"total":82,"votes":11441,"rating":1222.9282859359,"low":1216.2807943705886,"high":1229.5757775012114,"snapshot":"2026-10-07","attribution":"Blind human votes from Arena (https://arena.ai), CC BY 4.0. Arena checkpoint, not a quantized local-build score.","dataset":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset"},"sources":[{"label":"Qwen-Image-2.1 model card","url":"https://huggingface.co/Qwen/Qwen-Image-2.1","checkedAt":"2026-10-24","evidence":"With just **7B parameters in its visual generation component** (32 Single-Stream DiT layers)"},{"label":"Qwen-Image-2.1 model card, editing","url":"https://huggingface.co/Qwen/Qwen-Image-2.1","checkedAt":"2026-10-24","evidence":"Support up to **10 reference images**"},{"label":"Qwen-Image-2.1 license","url":"https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE","checkedAt":"2026-10-24","evidence":"FOR NON-COMMERCIAL PURPOSES ONLY."},{"label":"Qwen-Image-2.1 license, commercial use","url":"https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE","checkedAt":"2026-10-24","evidence":"You shall not use the Materials for any commercial purpose without obtaining a separate commercial license from us."},{"label":"mflux README","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/README.md","checkedAt":"2026-10-24","evidence":"Float16 compute for Apple M1-family GPUs on FLUX.2, FLUX 3 Image Klein, Z-Image and Qwen-Image 2.1"}]},{"id":"sd-3-5-large","name":"Stable Diffusion 3.5 Large","modality":"image","creator":"Stability AI","url":"https://modelfit.io/image/sd-3-5-large/","officialRepository":"https://huggingface.co/stabilityai/stable-diffusion-3.5-large","license":{"name":"Stability AI Community License","url":"https://stability.ai/community-license-agreement","commercial":"restricted","excludedTerritories":[],"note":"Commercial use requires registration with Stability AI. Licenses terminate above USD $1,000,000 in annual revenue until a separate license is requested."},"components":[{"id":"dit","name":"MM-DiT denoiser","role":"dit","stage":"denoise","required":true,"sources":[{"label":"Stable Diffusion 3.5 announcement","url":"https://stability.ai/news/introducing-stable-diffusion-3-5","checkedAt":"2026-10-24","evidence":"At 8.1 billion parameters"}],"parametersB":8.1},{"id":"clip_l","name":"CLIP-L text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"stable-diffusion.cpp SD3.5 guide","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/docs/sd3.md","checkedAt":"2026-10-24","evidence":"Download clip_l from"}]},{"id":"clip_g","name":"OpenCLIP bigG text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"stable-diffusion.cpp SD3.5 guide","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/docs/sd3.md","checkedAt":"2026-10-24","evidence":"Download clip_g from"}]},{"id":"t5xxl","name":"T5-XXL text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"stable-diffusion.cpp SD3.5 guide","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/docs/sd3.md","checkedAt":"2026-10-24","evidence":"Download t5xxl from"}]},{"id":"vae","name":"VAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"Stability AI SD3.5 reference code","url":"https://github.com/Stability-AI/sd3.5/blob/main/README.md","checkedAt":"2026-10-24","evidence":"the VAE Decoder (similar to previous SD models, but 16-channels and no postquantconv step)"}]}],"builds":[{"id":"original-comfyui","label":"Original safetensors","format":"safetensors","quantization":"Original weights","apps":["comfyui"],"files":[{"repo":"stabilityai/stable-diffusion-3.5-large","path":"sd3.5_large.safetensors","componentId":"dit","bytes":16460379262,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/clip_l.safetensors","componentId":"clip_l","bytes":246144152,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/clip_g.safetensors","componentId":"clip_g","bytes":1389382176,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/t5xxl_fp16.safetensors","componentId":"t5xxl","bytes":9787841024,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":167666902,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"}],"sources":[{"label":"SD 3.5 Large model card","url":"https://huggingface.co/stabilityai/stable-diffusion-3.5-large/resolve/main/README.md","checkedAt":"2026-10-24","evidence":"we recommend [ComfyUI](https://github.com/comfyanonymous/ComfyUI) for node-based UI inference"}]},{"id":"rapid-mlx-4bit","label":"Rapid-MLX 4-bit","format":"mlx","quantization":"4-bit","apps":["rapid-mlx"],"files":[{"repo":"argmaxinc/mlx-stable-diffusion-3.5-large-4bit-quantized","path":"sd3.5_large_4bit_quantized.safetensors","componentId":"dit","bytes":4951606576,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/argmaxinc/mlx-stable-diffusion-3.5-large-4bit-quantized?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/clip_l.safetensors","componentId":"clip_l","bytes":246144152,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/clip_g.safetensors","componentId":"clip_g","bytes":1389382176,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/t5xxl_fp16.safetensors","componentId":"t5xxl","bytes":9787841024,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":167666902,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"}],"sources":[{"label":"Rapid-MLX SD 3.5 Large dogfood","url":"https://github.com/raullenchai/Rapid-MLX/blob/dc60da43b612d70ffe29c5b6c6572fceb3cd67ab/docs/engineering/performance/2026-09-05-sd35-large-dogfood.md","checkedAt":"2026-10-24","evidence":"Primary checkpoint revision: `0f92f6c2a9f9e1abc6738209e87ac22b049a7d26`"},{"label":"Rapid-MLX SD 3.5 Large dogfood, encoders","url":"https://github.com/raullenchai/Rapid-MLX/blob/dc60da43b612d70ffe29c5b6c6572fceb3cd67ab/docs/engineering/performance/2026-09-05-sd35-large-dogfood.md","checkedAt":"2026-10-24","evidence":"uses all three published"},{"label":"DiffusionKit 4-bit checkpoint card","url":"https://huggingface.co/argmaxinc/mlx-stable-diffusion-3.5-large-4bit-quantized","checkedAt":"2026-10-24","evidence":"# Stable Diffusion 3.5 Large Quantized on DiffusionKit MLX!"}]},{"id":"gguf-q8_0","label":"Q8_0 GGUF","format":"gguf","quantization":"Q8_0","apps":["comfyui"],"files":[{"repo":"city96/stable-diffusion-3.5-large-gguf","path":"sd3.5_large-Q8_0.gguf","componentId":"dit","bytes":8779212512,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/city96/stable-diffusion-3.5-large-gguf?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/clip_l.safetensors","componentId":"clip_l","bytes":246144152,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/clip_g.safetensors","componentId":"clip_g","bytes":1389382176,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/t5xxl_fp16.safetensors","componentId":"t5xxl","bytes":9787841024,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":167666902,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"}],"sources":[{"label":"SD 3.5 Large GGUF card","url":"https://huggingface.co/city96/stable-diffusion-3.5-large-gguf","checkedAt":"2026-10-24","evidence":"The model files can be used with the [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF) custom node."},{"label":"ComfyUI-GGUF README","url":"https://github.com/city96/ComfyUI-GGUF/blob/6ea2651e7df66d7585f6ffee804b20e92fb38b8a/README.md","checkedAt":"2026-10-24","evidence":"- [stable-diffusion-3.5-large GGUF](https://huggingface.co/city96/stable-diffusion-3.5-large-gguf)"}]},{"id":"gguf-q4_0","label":"Q4_0 GGUF","format":"gguf","quantization":"Q4_0","apps":["comfyui"],"files":[{"repo":"city96/stable-diffusion-3.5-large-gguf","path":"sd3.5_large-Q4_0.gguf","componentId":"dit","bytes":4772054752,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/city96/stable-diffusion-3.5-large-gguf?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/clip_l.safetensors","componentId":"clip_l","bytes":246144152,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/clip_g.safetensors","componentId":"clip_g","bytes":1389382176,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"text_encoders/t5xxl_fp16.safetensors","componentId":"t5xxl","bytes":9787841024,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"},{"repo":"stabilityai/stable-diffusion-3.5-large","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":167666902,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-3.5-large?blobs=true"}],"sources":[{"label":"SD 3.5 Large GGUF card","url":"https://huggingface.co/city96/stable-diffusion-3.5-large-gguf","checkedAt":"2026-10-24","evidence":"The model files can be used with the [ComfyUI-GGUF](https://github.com/city96/ComfyUI-GGUF) custom node."},{"label":"ComfyUI-GGUF README","url":"https://github.com/city96/ComfyUI-GGUF/blob/6ea2651e7df66d7585f6ffee804b20e92fb38b8a/README.md","checkedAt":"2026-10-24","evidence":"- [stable-diffusion-3.5-large GGUF](https://huggingface.co/city96/stable-diffusion-3.5-large-gguf)"}]}],"publishedMeasurements":[{"id":"sd-3-5-large-rapid-mlx-4bit-m3-ultra-256gb-512","modelId":"sd-3-5-large","buildId":"rapid-mlx-4bit","appId":"rapid-mlx","machine":{"chip":"Apple M3 Ultra","ramGb":256},"workload":{"resolution":"512x512","steps":28,"quantization":"4-bit","notes":"Seed 42, MLX 0.32.1, macOS 26.5.2, second clean-process run; the first run took 62.17 s. Peak is the OS peak memory footprint, with 0 swaps. Maximum resident set size was lower."},"outcome":"success","timeSeconds":57.29,"peakMemoryGb":16.667,"peakMemoryText":"16,667,122,480 bytes","publishedAt":"2026-09-05","grade":"A","source":{"label":"Rapid-MLX SD 3.5 Large dogfood","url":"https://github.com/raullenchai/Rapid-MLX/blob/dc60da43b612d70ffe29c5b6c6572fceb3cd67ab/docs/engineering/performance/2026-09-05-sd35-large-dogfood.md","checkedAt":"2026-10-24","evidence":"- Apple M3 Ultra, 256 GiB unified memory"}}],"quality":{"modelId":"sd-3-5-large","category":"text_to_image","sourceName":"stable-diffusion-v35-large","rank":81,"total":82,"votes":23396,"rating":938.2718023761549,"low":933.7579623497016,"high":942.7856424026081,"snapshot":"2026-10-07","attribution":"Blind human votes from Arena (https://arena.ai), CC BY 4.0. Arena checkpoint, not a quantized local-build score.","dataset":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset"},"sources":[{"label":"Stability AI Community License","url":"https://stability.ai/community-license-agreement","checkedAt":"2026-10-24","evidence":"generate more than USD $1,000,000 in annual revenue"},{"label":"stable-diffusion.cpp README","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"[SD3/SD3.5](./docs/sd3.md)"}]},{"id":"sdxl","name":"SDXL Base 1.0","modality":"image","creator":"Stability AI","url":"https://modelfit.io/image/sdxl/","officialRepository":"https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0","license":{"name":"CreativeML Open RAIL++-M License","url":"https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENSE.md","commercial":"yes","excludedTerritories":[],"note":"Broad use grant with the use-based restrictions in Attachment A."},"components":[{"id":"unet","name":"SDXL UNet","role":"dit","stage":"denoise","required":true,"sources":[{"label":"Rapid-MLX SDXL Base dogfood","url":"https://github.com/raullenchai/Rapid-MLX/blob/e8755711ffedc1365409c89909b6f958be51cded/docs/engineering/performance/2026-09-05-sdxl-base-dogfood.md","checkedAt":"2026-10-24","evidence":"the UNet is quantized to 8-bit at load"}]},{"id":"clip_l","name":"CLIP ViT-L text encoder (inside single-file checkpoints)","role":"text_encoder","stage":"text","required":false,"sources":[{"label":"SDXL Base model card","url":"https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0","checkedAt":"2026-10-24","evidence":"uses two fixed, pretrained text encoders"}]},{"id":"clip_g","name":"OpenCLIP ViT-bigG text encoder (inside single-file checkpoints)","role":"text_encoder","stage":"text","required":false,"sources":[{"label":"SDXL Base model card","url":"https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0","checkedAt":"2026-10-24","evidence":"uses two fixed, pretrained text encoders"}]},{"id":"vae","name":"VAE (inside single-file checkpoints)","role":"vae","stage":"decode","required":false,"sources":[{"label":"SDXL Base model card","url":"https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0","checkedAt":"2026-10-24","evidence":"vae=base.vae,"}]}],"builds":[{"id":"original-checkpoint","label":"Original single-file checkpoint","format":"safetensors","quantization":"Original weights","apps":["comfyui"],"files":[{"repo":"stabilityai/stable-diffusion-xl-base-1.0","path":"sd_xl_base_1.0.safetensors","componentId":"unet","bytes":6938078334,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-xl-base-1.0?blobs=true"}],"sources":[{"label":"ComfyUI README","url":"https://github.com/Comfy-Org/ComfyUI/blob/7cebca52227c63b4180ceed448cbd7ad22903b6d/README.md","checkedAt":"2026-10-24","evidence":"Put your SD checkpoints (the huge ckpt/safetensors files) in: models/checkpoints"}]},{"id":"rapid-mlx-8bit-unet","label":"Rapid-MLX 8-bit UNet","format":"safetensors","quantization":"8-bit UNet (quantized at load)","apps":["rapid-mlx"],"files":[{"repo":"stabilityai/stable-diffusion-xl-base-1.0","path":"unet/diffusion_pytorch_model.fp16.safetensors","componentId":"unet","bytes":5135149760,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-xl-base-1.0?blobs=true"},{"repo":"stabilityai/stable-diffusion-xl-base-1.0","path":"text_encoder/model.fp16.safetensors","componentId":"clip_l","bytes":246144152,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-xl-base-1.0?blobs=true"},{"repo":"stabilityai/stable-diffusion-xl-base-1.0","path":"text_encoder_2/model.fp16.safetensors","componentId":"clip_g","bytes":1389382176,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-xl-base-1.0?blobs=true"},{"repo":"stabilityai/stable-diffusion-xl-base-1.0","path":"vae/diffusion_pytorch_model.fp16.safetensors","componentId":"vae","bytes":167335342,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/stabilityai/stable-diffusion-xl-base-1.0?blobs=true"}],"sources":[{"label":"Rapid-MLX SDXL Base dogfood","url":"https://github.com/raullenchai/Rapid-MLX/blob/e8755711ffedc1365409c89909b6f958be51cded/docs/engineering/performance/2026-09-05-sdxl-base-dogfood.md","checkedAt":"2026-10-24","evidence":"the UNet is quantized to 8-bit at load"},{"label":"Rapid-MLX SDXL Base dogfood, checkpoint","url":"https://github.com/raullenchai/Rapid-MLX/blob/e8755711ffedc1365409c89909b6f958be51cded/docs/engineering/performance/2026-09-05-sdxl-base-dogfood.md","checkedAt":"2026-10-24","evidence":"Model `stabilityai/stable-diffusion-xl-base-1.0`"}]},{"id":"gguf-q8_0","label":"Q8_0 GGUF","format":"gguf","quantization":"Q8_0","apps":["stable-diffusion-cpp"],"files":[{"repo":"kostakoff/stable-diffusion-xl-base-1.0-GGUF","path":"sd_xl_base_1.0_0_Q8_0.gguf","componentId":"unet","bytes":4180186816,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/kostakoff/stable-diffusion-xl-base-1.0-GGUF?blobs=true"}],"sources":[{"label":"SDXL GGUF card","url":"https://huggingface.co/kostakoff/stable-diffusion-xl-base-1.0-GGUF","checkedAt":"2026-10-24","evidence":"Quantized versions of [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) in **GGUF** format for use with [stable-diffusion.cpp](https://github.com/leejet/stable-diffusion.cpp)."},{"label":"SDXL GGUF card, conversion","url":"https://huggingface.co/kostakoff/stable-diffusion-xl-base-1.0-GGUF","checkedAt":"2026-10-24","evidence":"Converted from the original [sd_xl_base_1.0_0.9vae.safetensors]"}]},{"id":"gguf-q4_k","label":"Q4_K GGUF","format":"gguf","quantization":"Q4_K","apps":["stable-diffusion-cpp"],"files":[{"repo":"kostakoff/stable-diffusion-xl-base-1.0-GGUF","path":"sd_xl_base_1.0_0_Q4_K.gguf","componentId":"unet","bytes":2797752960,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/kostakoff/stable-diffusion-xl-base-1.0-GGUF?blobs=true"}],"sources":[{"label":"SDXL GGUF card","url":"https://huggingface.co/kostakoff/stable-diffusion-xl-base-1.0-GGUF","checkedAt":"2026-10-24","evidence":"Quantized versions of [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) in **GGUF** format for use with [stable-diffusion.cpp](https://github.com/leejet/stable-diffusion.cpp)."},{"label":"SDXL GGUF card, conversion","url":"https://huggingface.co/kostakoff/stable-diffusion-xl-base-1.0-GGUF","checkedAt":"2026-10-24","evidence":"Converted from the original [sd_xl_base_1.0_0.9vae.safetensors]"}]}],"publishedMeasurements":[{"id":"sdxl-rapid-mlx-8bit-unet-m3-ultra-256gb-1024","modelId":"sdxl","buildId":"rapid-mlx-8bit-unet","appId":"rapid-mlx","machine":{"chip":"Apple M3 Ultra","ramGb":256},"workload":{"resolution":"1024x1024","steps":30,"quantization":"8-bit UNet (quantized at load)","notes":"Guidance 5.0, seed 42, tiled VAE, macOS 26.5.2, MLX 0.32.2. Time is cold model construction plus generation. Peak is MLX peak allocation."},"outcome":"success","timeSeconds":61.72,"peakMemoryGb":13.336,"peakMemoryText":"12.42 GiB","publishedAt":"2026-09-05","grade":"A","source":{"label":"Rapid-MLX SDXL Base dogfood","url":"https://github.com/raullenchai/Rapid-MLX/blob/e8755711ffedc1365409c89909b6f958be51cded/docs/engineering/performance/2026-09-05-sdxl-base-dogfood.md","checkedAt":"2026-10-24","evidence":"- Apple M3 Ultra, 256 GiB unified memory"}}],"quality":null,"sources":[{"label":"SDXL Base license","url":"https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENSE.md","checkedAt":"2026-10-24","evidence":"CreativeML Open RAIL++-M License dated July 26, 2023"},{"label":"SDXL Base model card, standalone","url":"https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0","checkedAt":"2026-10-24","evidence":"the base model can be used as a standalone module."},{"label":"Draw Things App Store listing data","url":"https://itunes.apple.com/lookup?id=6444050820&country=us","checkedAt":"2026-10-24","evidence":"Harness the latest models like SDXL, FLUX.1, Wan 2.2, Qwen Image, Z Image, FLUX.2 and LTX-2.3"}]},{"id":"wan-2-2-5b","name":"Wan2.2 TI2V 5B","modality":"video","creator":"Wan-AI","url":"https://modelfit.io/video/wan-2-2-5b/","officialRepository":"https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B","license":{"name":"Apache License 2.0","url":"https://github.com/Wan-Video/Wan2.2/blob/main/LICENSE.txt","commercial":"yes","excludedTerritories":[]},"components":[{"id":"dit","name":"TI2V-5B dense transformer","role":"dit","stage":"denoise","required":true,"parametersB":5,"sources":[{"label":"Wan2.2 README","url":"https://github.com/Wan-Video/Wan2.2/blob/main/README.md","checkedAt":"2026-10-24","evidence":"a 5B dense model, i.e., TI2V-5B, is released"}]},{"id":"text_encoder","name":"UMT5-XXL text encoder","role":"text_encoder","stage":"text","required":true,"sources":[]},{"id":"video_vae","name":"Wan2.2-VAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"Wan2.2 README","url":"https://github.com/Wan-Video/Wan2.2/blob/main/README.md","checkedAt":"2026-10-24","evidence":"compression ratio of $4\\times16\\times16$"}]}],"builds":[{"id":"official-unquantized","label":"Official Wan-AI weights (unquantized)","format":"safetensors","quantization":"unquantized","apps":["mlx-video"],"files":[{"repo":"Wan-AI/Wan2.2-TI2V-5B","path":"diffusion_pytorch_model-00001-of-00003.safetensors","componentId":"dit","bytes":9825014472,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Wan-AI/Wan2.2-TI2V-5B?blobs=true"},{"repo":"Wan-AI/Wan2.2-TI2V-5B","path":"diffusion_pytorch_model-00002-of-00003.safetensors","componentId":"dit","bytes":9995661736,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Wan-AI/Wan2.2-TI2V-5B?blobs=true"},{"repo":"Wan-AI/Wan2.2-TI2V-5B","path":"diffusion_pytorch_model-00003-of-00003.safetensors","componentId":"dit","bytes":178558176,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Wan-AI/Wan2.2-TI2V-5B?blobs=true"},{"repo":"Wan-AI/Wan2.2-TI2V-5B","path":"models_t5_umt5-xxl-enc-bf16.pth","componentId":"text_encoder","bytes":11361920418,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Wan-AI/Wan2.2-TI2V-5B?blobs=true"},{"repo":"Wan-AI/Wan2.2-TI2V-5B","path":"Wan2.2_VAE.pth","componentId":"video_vae","bytes":2818839170,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Wan-AI/Wan2.2-TI2V-5B?blobs=true"}],"sources":[{"label":"Wan2.2 TI2V 5B model card","url":"https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B","checkedAt":"2026-10-24","evidence":"This model supports both text-to-video and image-to-video generation at 720P resolution with 24fps"},{"label":"mlx-video Wan README","url":"https://github.com/Blaizzy/mlx-video/blob/main/mlx_video/models/wan_2/README.md","checkedAt":"2026-10-24","evidence":"| **Sizes** | 1.3B, 14B | 14B | 14B | 5B |"}]},{"id":"fastmetal-5b-qad","label":"FastMetal 5B QAD (DMD2-distilled INT8 derivative, not the base model)","format":"mlx","quantization":"INT8 3-step DMD2","apps":["fastvideo"],"files":[{"repo":"FastVideo/FastMetal-5B-QAD","path":"mlx_dit.safetensors","componentId":"dit","bytes":5314678720,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/FastVideo/FastMetal-5B-QAD?blobs=true"},{"repo":"FastVideo/FastMetal-5B-QAD","path":"text_encoder/model-00001-of-00003.safetensors","componentId":"text_encoder","bytes":4935812536,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/FastVideo/FastMetal-5B-QAD?blobs=true"},{"repo":"FastVideo/FastMetal-5B-QAD","path":"text_encoder/model-00002-of-00003.safetensors","componentId":"text_encoder","bytes":4983103192,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/FastVideo/FastMetal-5B-QAD?blobs=true"},{"repo":"FastVideo/FastMetal-5B-QAD","path":"text_encoder/model-00003-of-00003.safetensors","componentId":"text_encoder","bytes":1442935480,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/FastVideo/FastMetal-5B-QAD?blobs=true"},{"repo":"FastVideo/FastMetal-5B-QAD","path":"vae/diffusion_pytorch_model.safetensors","componentId":"video_vae","bytes":2818777808,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/FastVideo/FastMetal-5B-QAD?blobs=true"}],"sources":[{"label":"FastMetal-5B-QAD card","url":"https://huggingface.co/FastVideo/FastMetal-5B-QAD","checkedAt":"2026-10-24","evidence":"a DMD2-distilled Wan2.2 TI2V 5B with a"},{"label":"FastMetal-5B-QAD card","url":"https://huggingface.co/FastVideo/FastMetal-5B-QAD","checkedAt":"2026-10-24","evidence":"| Quantization | affine INT8, group size 64, QAT-trained |"}]},{"id":"comfyui-fp16","label":"Comfy-Org repackaged FP16 with FP8 scaled UMT5","format":"safetensors","quantization":"FP16","apps":["comfyui"],"files":[{"repo":"Comfy-Org/Wan-2.2","path":"split_files/diffusion_models/wan2.2_ti2v_5B_fp16.safetensors","componentId":"dit","bytes":9999658848,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors","componentId":"text_encoder","bytes":6735906897,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/vae/wan2.2_vae.safetensors","componentId":"video_vae","bytes":1409400960,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"}],"sources":[{"label":"Comfy-Org Wan 2.2 repackaged card","url":"https://huggingface.co/Comfy-Org/Wan-2.2","checkedAt":"2026-10-24","evidence":"wan2.2_ti2v_5B_fp16.safetensors"}]},{"id":"gguf-q8-0","label":"QuantStack GGUF Q8_0","format":"gguf","quantization":"Q8_0","apps":["comfyui"],"files":[{"repo":"QuantStack/Wan2.2-TI2V-5B-GGUF","path":"Wan2.2-TI2V-5B-Q8_0.gguf","componentId":"dit","bytes":5400179040,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/QuantStack/Wan2.2-TI2V-5B-GGUF?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors","componentId":"text_encoder","bytes":6735906897,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/vae/wan2.2_vae.safetensors","componentId":"video_vae","bytes":1409400960,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"}],"sources":[{"label":"QuantStack Wan2.2 TI2V 5B GGUF card","url":"https://huggingface.co/QuantStack/Wan2.2-TI2V-5B-GGUF","checkedAt":"2026-10-24","evidence":"The model can be used with the ComfyUI custom node"},{"label":"Comfy-Org Wan 2.2 repackaged card","url":"https://huggingface.co/Comfy-Org/Wan-2.2","checkedAt":"2026-10-24","evidence":"wan2.2_vae.safetensors"}]}],"publishedMeasurements":[{"id":"wan-2-2-5b-mlx-video-m3max-64gb-i2v-121f","modelId":"wan-2-2-5b","buildId":"official-unquantized","appId":"mlx-video","machine":{"device":"MacBook Pro","chip":"Apple M3 Max","ramGb":64},"workload":{"resolution":"1280x704","frames":121,"fps":24,"durationSeconds":5,"steps":40,"quantization":"unquantized","notes":"I2V with an input image, 121 frames, 5 seconds as the author describes it. The run date is not stated; the date shown is the article date."},"outcome":"success","timeSeconds":5760,"timeText":"96 minutes","peakMemoryGb":57,"publishedAt":"2026-10-10","grade":"A","source":{"label":"rotecodefraktion: local Wan video on a Mac","url":"https://www.rotecodefraktion.de/en/blog/lokale-ki-video-wan-mac/","checkedAt":"2026-10-24","evidence":"Our Mac spent 96 minutes on this clip, with a peak memory of 57 GB."}},{"id":"wan-2-2-5b-mlx-video-m3max-64gb-t2v-41f","modelId":"wan-2-2-5b","buildId":"official-unquantized","appId":"mlx-video","machine":{"device":"MacBook Pro","chip":"Apple M3 Max","ramGb":64},"workload":{"resolution":"1280x704","frames":41,"fps":24,"durationSeconds":1.7,"steps":40,"quantization":"unquantized","notes":"T2V, 41 frames. Time is the article total: encoder load, transformer load, 40 diffusion steps and VAE decode. The run date is not stated; the date shown is the article date."},"outcome":"success","timeSeconds":1063,"timeText":"17:43","peakMemoryGb":56.4,"publishedAt":"2026-10-10","grade":"A","source":{"label":"rotecodefraktion: local Wan video on a Mac","url":"https://www.rotecodefraktion.de/en/blog/lokale-ki-video-wan-mac/","checkedAt":"2026-10-24","evidence":"17:43 min"}},{"id":"wan-2-2-5b-fastmetal-m4max-36gb","modelId":"wan-2-2-5b","buildId":"fastmetal-5b-qad","appId":"fastvideo","machine":{"chip":"Apple M4 Max","ramGb":36},"workload":{"resolution":"1280x704","frames":81,"steps":3,"quantization":"INT8 3-step DMD2","notes":"FastMetal-QAD derivative, INT8 DiT, TAEHV decode. End-to-end time includes prompt encoding, DiT load, denoising and decode/export. The date is the article date; the run date is not stated."},"outcome":"success","timeSeconds":151.42,"peakMemoryGb":10.03,"peakMemoryText":"9.34 GiB","publishedAt":"2026-08-10","grade":"A","source":{"label":"FastMetal-QAD announcement","url":"https://haoailab.com/blogs/fastmetal/","checkedAt":"2026-10-24","evidence":"5B 704×1280×81 151.42s 98.50s 9.34 GiB"}},{"id":"wan-2-2-5b-fastmetal-m5air-24gb","modelId":"wan-2-2-5b","buildId":"fastmetal-5b-qad","appId":"fastvideo","machine":{"device":"13-inch MacBook Air","chip":"Apple M5","ramGb":24},"workload":{"resolution":"1280x704","frames":81,"steps":3,"quantization":"INT8 3-step DMD2","notes":"FastMetal-QAD derivative, Baseline mode, average over repeated runs. The date is the article date; the run dates are not stated."},"outcome":"success","timeSeconds":200.1,"peakMemoryGb":10.24,"peakMemoryText":"9.54 GiB","publishedAt":"2026-08-10","grade":"A","source":{"label":"FastMetal-QAD announcement","url":"https://haoailab.com/blogs/fastmetal/","checkedAt":"2026-10-24","evidence":"5B Baseline 200.1s 9.54 GiB"}}],"quality":null,"sources":[{"label":"Wan2.2 TI2V 5B model card","url":"https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B","checkedAt":"2026-10-24","evidence":"This model supports both text-to-video and image-to-video generation at 720P resolution with 24fps"},{"label":"Wan2.2 README","url":"https://github.com/Wan-Video/Wan2.2/blob/main/README.md","checkedAt":"2026-10-24","evidence":"The models in this repository are licensed under the Apache 2.0 License."},{"label":"Apache License 2.0","url":"https://github.com/Wan-Video/Wan2.2/blob/main/LICENSE.txt","checkedAt":"2026-10-24","evidence":"Version 2.0, January 2004"}]},{"id":"wan-2-2-a14b","name":"Wan2.2 A14B","modality":"video","creator":"Wan-AI","url":"https://modelfit.io/video/wan-2-2-a14b/","officialRepository":"https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B","license":{"name":"Apache License 2.0","url":"https://github.com/Wan-Video/Wan2.2/blob/main/LICENSE.txt","commercial":"yes","excludedTerritories":[]},"components":[{"id":"dit_high","name":"High-noise expert","role":"dit","stage":"denoise","required":true,"parametersB":14,"sources":[{"label":"Wan2.2 README","url":"https://github.com/Wan-Video/Wan2.2/blob/main/README.md","checkedAt":"2026-10-24","evidence":"a high-noise expert for the early stages, focusing on overall layout"}]},{"id":"dit_low","name":"Low-noise expert","role":"dit","stage":"denoise","required":true,"parametersB":14,"sources":[{"label":"Wan2.2 README","url":"https://github.com/Wan-Video/Wan2.2/blob/main/README.md","checkedAt":"2026-10-24","evidence":"a low-noise expert for the later stages, refining video details"}]},{"id":"text_encoder","name":"UMT5-XXL text encoder","role":"text_encoder","stage":"text","required":true,"sources":[]},{"id":"video_vae","name":"Wan2.1 VAE","role":"vae","stage":"decode","required":true,"sources":[]}],"builds":[{"id":"pocketai-mlx-q4","label":"PocketAI MLX Q4 (group size 64)","format":"mlx","quantization":"Q4","apps":["pocketai-mlx"],"files":[{"repo":"PocketAiHub/Wan2.2-T2V-A14B-MLX","path":"q4/high_noise_model.safetensors","componentId":"dit_high","bytes":8378982847,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/PocketAiHub/Wan2.2-T2V-A14B-MLX?blobs=true"},{"repo":"PocketAiHub/Wan2.2-T2V-A14B-MLX","path":"q4/low_noise_model.safetensors","componentId":"dit_low","bytes":8378982717,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/PocketAiHub/Wan2.2-T2V-A14B-MLX?blobs=true"},{"repo":"PocketAiHub/Wan2.2-T2V-A14B-MLX","path":"common/t5_encoder.safetensors","componentId":"text_encoder","bytes":11361845591,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/PocketAiHub/Wan2.2-T2V-A14B-MLX?blobs=true"},{"repo":"PocketAiHub/Wan2.2-T2V-A14B-MLX","path":"common/vae.safetensors","componentId":"video_vae","bytes":507591266,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/PocketAiHub/Wan2.2-T2V-A14B-MLX?blobs=true"}],"sources":[{"label":"PocketAI Wan2.2 T2V A14B MLX card","url":"https://huggingface.co/PocketAiHub/Wan2.2-T2V-A14B-MLX","checkedAt":"2026-10-24","evidence":"Apple-Silicon-native Q4 MLX packaging of Wan2.2 T2V A14B"}]},{"id":"comfyui-fp8-scaled","label":"Comfy-Org repackaged FP8 scaled","format":"safetensors","quantization":"FP8 scaled","apps":["comfyui"],"files":[{"repo":"Comfy-Org/Wan-2.2","path":"split_files/diffusion_models/wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors","componentId":"dit_high","bytes":14293923632,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/diffusion_models/wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors","componentId":"dit_low","bytes":14293923632,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors","componentId":"text_encoder","bytes":6735906897,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/vae/wan_2.1_vae.safetensors","componentId":"video_vae","bytes":253815318,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"}],"sources":[{"label":"Comfy-Org Wan 2.2 repackaged card","url":"https://huggingface.co/Comfy-Org/Wan-2.2","checkedAt":"2026-10-24","evidence":"wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors"},{"label":"Comfy-Org Wan 2.2 repackaged card","url":"https://huggingface.co/Comfy-Org/Wan-2.2","checkedAt":"2026-10-24","evidence":"umt5_xxl_fp8_e4m3fn_scaled.safetensors"}]},{"id":"gguf-q4-k-m","label":"QuantStack GGUF Q4_K_M (high and low noise)","format":"gguf","quantization":"Q4_K_M","apps":["comfyui"],"files":[{"repo":"QuantStack/Wan2.2-T2V-A14B-GGUF","path":"HighNoise/Wan2.2-T2V-A14B-HighNoise-Q4_K_M.gguf","componentId":"dit_high","bytes":9650090496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/QuantStack/Wan2.2-T2V-A14B-GGUF?blobs=true"},{"repo":"QuantStack/Wan2.2-T2V-A14B-GGUF","path":"LowNoise/Wan2.2-T2V-A14B-LowNoise-Q4_K_M.gguf","componentId":"dit_low","bytes":9650090496,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/QuantStack/Wan2.2-T2V-A14B-GGUF?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors","componentId":"text_encoder","bytes":6735906897,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"},{"repo":"Comfy-Org/Wan-2.2","path":"split_files/vae/wan_2.1_vae.safetensors","componentId":"video_vae","bytes":253815318,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Comfy-Org/Wan-2.2?blobs=true"}],"sources":[{"label":"QuantStack Wan2.2 T2V A14B GGUF card","url":"https://huggingface.co/QuantStack/Wan2.2-T2V-A14B-GGUF","checkedAt":"2026-10-24","evidence":"The model can be used with the ComfyUI custom node"},{"label":"Comfy-Org Wan 2.2 repackaged card","url":"https://huggingface.co/Comfy-Org/Wan-2.2","checkedAt":"2026-10-24","evidence":"umt5_xxl_fp8_e4m3fn_scaled.safetensors"}]}],"publishedMeasurements":[{"id":"wan-2-2-a14b-pocketai-q4-m5max-128gb","modelId":"wan-2-2-a14b","buildId":"pocketai-mlx-q4","appId":"pocketai-mlx","machine":{"chip":"Apple M5 Max","ramGb":128},"workload":{"resolution":"512x320","frames":81,"fps":16,"durationSeconds":5.0625,"steps":10,"quantization":"Q4","notes":"UniPC scheduler, guidance 3/4, identical prompt and seed. Time is end to end; peak is maximum process RSS. The repository carries no run date, so the date shown is its last-modified date."},"outcome":"success","timeSeconds":625,"timeText":"10:25","peakMemoryGb":17.58,"peakMemoryText":"16.37 GiB","publishedAt":"2026-08-15","grade":"A","source":{"label":"PocketAI Wan2.2 T2V A14B MLX card","url":"https://huggingface.co/PocketAiHub/Wan2.2-T2V-A14B-MLX","checkedAt":"2026-10-24","evidence":"| Q4 | 10:25 | 9:06 | 16.37 GiB |"}}],"quality":{"modelId":"wan-2-2-a14b","category":"text_to_video","sourceName":"wan-v2.2-a14b","rank":42,"total":48,"votes":10458,"rating":1131.564430836072,"low":1116.4282381134221,"high":1146.7006235587219,"snapshot":"2026-09-22","attribution":"Blind human votes from Arena (https://arena.ai), CC BY 4.0. Arena checkpoint, not a quantized local-build score.","dataset":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset"},"sources":[{"label":"Wan2.2 T2V A14B model card","url":"https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B","checkedAt":"2026-10-24","evidence":"This repository contains our T2V-A14B model, which supports generating 5s videos at both 480P and 720P resolutions."},{"label":"Wan2.2 README","url":"https://github.com/Wan-Video/Wan2.2/blob/main/README.md","checkedAt":"2026-10-24","evidence":"The models in this repository are licensed under the Apache 2.0 License."},{"label":"Apache License 2.0","url":"https://github.com/Wan-Video/Wan2.2/blob/main/LICENSE.txt","checkedAt":"2026-10-24","evidence":"Version 2.0, January 2004"}]},{"id":"z-image-turbo","name":"Z-Image-Turbo","modality":"image","creator":"Tongyi-MAI","url":"https://modelfit.io/image/z-image-turbo/","officialRepository":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo","license":{"name":"Apache 2.0","url":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo","commercial":"yes","excludedTerritories":[]},"components":[{"id":"dit","name":"Z-Image transformer","role":"dit","stage":"denoise","required":true,"sources":[{"label":"Z-Image-Turbo pipeline config","url":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo/blob/main/model_index.json","checkedAt":"2026-10-24","evidence":"ZImageTransformer2DModel"}]},{"id":"text_encoder","name":"Qwen3 text encoder","role":"text_encoder","stage":"text","required":true,"sources":[{"label":"Z-Image-Turbo pipeline config","url":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo/blob/main/model_index.json","checkedAt":"2026-10-24","evidence":"Qwen3Model"}]},{"id":"vae","name":"VAE","role":"vae","stage":"decode","required":true,"sources":[{"label":"Z-Image-Turbo pipeline config","url":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo/blob/main/model_index.json","checkedAt":"2026-10-24","evidence":"AutoencoderKL"}]}],"builds":[{"id":"original-safetensors","label":"Original safetensors","format":"safetensors","quantization":"Original weights","apps":["diffusers"],"files":[{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"transformer/diffusion_pytorch_model-00001-of-00003.safetensors","componentId":"dit","bytes":9973693184,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"transformer/diffusion_pytorch_model-00002-of-00003.safetensors","componentId":"dit","bytes":9973714824,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"transformer/diffusion_pytorch_model-00003-of-00003.safetensors","componentId":"dit","bytes":4672282880,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00001-of-00003.safetensors","componentId":"text_encoder","bytes":3957900840,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00002-of-00003.safetensors","componentId":"text_encoder","bytes":3987450520,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00003-of-00003.safetensors","componentId":"text_encoder","bytes":99630640,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":167666902,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"}],"sources":[{"label":"Z-Image-Turbo model card","url":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo","checkedAt":"2026-10-24","evidence":"from diffusers import ZImagePipeline"}]},{"id":"mflux-4bit","label":"mflux 4-bit","format":"safetensors","quantization":"4-bit (mflux --quantize 4)","apps":["mflux"],"files":[{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"transformer/diffusion_pytorch_model-00001-of-00003.safetensors","componentId":"dit","bytes":9973693184,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"transformer/diffusion_pytorch_model-00002-of-00003.safetensors","componentId":"dit","bytes":9973714824,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"transformer/diffusion_pytorch_model-00003-of-00003.safetensors","componentId":"dit","bytes":4672282880,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00001-of-00003.safetensors","componentId":"text_encoder","bytes":3957900840,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00002-of-00003.safetensors","componentId":"text_encoder","bytes":3987450520,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00003-of-00003.safetensors","componentId":"text_encoder","bytes":99630640,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":167666902,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"}],"sources":[{"label":"mflux model registry","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/src/mflux/models/common/config/model_config.py","checkedAt":"2026-10-24","evidence":"model_name=\"Tongyi-MAI/Z-Image-Turbo\","},{"label":"mflux README","url":"https://github.com/mflux-community/mflux/blob/a3b21fcaa88726b2267bd42bfc2e2d722f7d5a47/README.md","checkedAt":"2026-10-24","evidence":"[Z-Image](src/mflux/models/z_image/README.md)"}]},{"id":"gguf-q8_0","label":"Q8_0 GGUF","format":"gguf","quantization":"Q8_0","apps":["stable-diffusion-cpp"],"files":[{"repo":"unsloth/Z-Image-Turbo-GGUF","path":"z-image-turbo-Q8_0.gguf","componentId":"dit","bytes":7224707136,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/Z-Image-Turbo-GGUF?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00001-of-00003.safetensors","componentId":"text_encoder","bytes":3957900840,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00002-of-00003.safetensors","componentId":"text_encoder","bytes":3987450520,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00003-of-00003.safetensors","componentId":"text_encoder","bytes":99630640,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":167666902,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"}],"sources":[{"label":"Z-Image-Turbo GGUF card","url":"https://huggingface.co/unsloth/Z-Image-Turbo-GGUF","checkedAt":"2026-10-24","evidence":"This is a GGUF quantized version of [Z-Image-Turbo]"},{"label":"stable-diffusion.cpp README","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"[Z-Image](./docs/z_image.md)"}]},{"id":"gguf-q4_k_m","label":"Q4_K_M GGUF","format":"gguf","quantization":"Q4_K_M","apps":["stable-diffusion-cpp"],"files":[{"repo":"unsloth/Z-Image-Turbo-GGUF","path":"z-image-turbo-Q4_K_M.gguf","componentId":"dit","bytes":5017613376,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/unsloth/Z-Image-Turbo-GGUF?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00001-of-00003.safetensors","componentId":"text_encoder","bytes":3957900840,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00002-of-00003.safetensors","componentId":"text_encoder","bytes":3987450520,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"text_encoder/model-00003-of-00003.safetensors","componentId":"text_encoder","bytes":99630640,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"},{"repo":"Tongyi-MAI/Z-Image-Turbo","path":"vae/diffusion_pytorch_model.safetensors","componentId":"vae","bytes":167666902,"probedAt":"2026-10-24","sourceUrl":"https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?blobs=true"}],"sources":[{"label":"Z-Image-Turbo GGUF card","url":"https://huggingface.co/unsloth/Z-Image-Turbo-GGUF","checkedAt":"2026-10-24","evidence":"This is a GGUF quantized version of [Z-Image-Turbo]"},{"label":"stable-diffusion.cpp README","url":"https://github.com/leejet/stable-diffusion.cpp/blob/95ff76230889898e1bf40058ba272a3699b50dc6/README.md","checkedAt":"2026-10-24","evidence":"[Z-Image](./docs/z_image.md)"}]}],"publishedMeasurements":[{"id":"z-image-turbo-mflux-4bit-m1-max-64gb-1024","modelId":"z-image-turbo","buildId":"mflux-4bit","appId":"mflux","machine":{"chip":"Apple M1 Max","ramGb":64},"workload":{"resolution":"1024x1024","steps":8,"quantization":"4-bit (mflux --quantize 4)","notes":"Seed 1, macOS 26, MLX 0.32.3, on AC power. Run before mflux changed its default float32 stream. Two runs took 102.4 s and 94.8 s; the slower is shown."},"outcome":"success","timeSeconds":102.4,"peakMemoryGb":26.13,"publishedAt":"2026-10-03","grade":"A","source":{"label":"mflux pull request 820","url":"https://github.com/mflux-community/mflux/pull/820","checkedAt":"2026-10-24","evidence":"Real images, M1 Max 64 GB, macOS 26, MLX 0.32.3, on AC power."}}],"quality":{"modelId":"z-image-turbo","category":"text_to_image","sourceName":"z-image-turbo","rank":59,"total":82,"votes":24808,"rating":1083.4963498301272,"low":1078.1954990327044,"high":1088.7972006275502,"snapshot":"2026-10-07","attribution":"Blind human votes from Arena (https://arena.ai), CC BY 4.0. Arena checkpoint, not a quantized local-build score.","dataset":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset"},"sources":[{"label":"Z-Image-Turbo model card","url":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo","checkedAt":"2026-10-24","evidence":"Z-Image is a powerful and highly efficient image generation model family with **6B** parameters."},{"label":"Z-Image-Turbo model card, example","url":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo","checkedAt":"2026-10-24","evidence":"num_inference_steps=9,  # This actually results in 8 DiT forwards"},{"label":"Z-Image-Turbo model card, license","url":"https://huggingface.co/Tongyi-MAI/Z-Image-Turbo","checkedAt":"2026-10-24","evidence":"license: apache-2.0"},{"label":"Draw Things App Store listing data","url":"https://itunes.apple.com/lookup?id=6444050820&country=us","checkedAt":"2026-10-24","evidence":"Harness the latest models like SDXL, FLUX.1, Wan 2.2, Qwen Image, Z Image, FLUX.2 and LTX-2.3"}]}]}