Files read on Hugging Face: 2026-10-24

Stable Diffusion 3.5 Large on Mac: files, RAM and published times

8.1 billion parameter MM-DiT text-to-image model with three text encoders, positioned for professional use at 1 megapixel. The Stability AI Community License requires registration for commercial use and ends above USD $1,000,000 annual revenue.

Stability AI · Updated 2026-10-24
Answer

Stable Diffusion 3.5 Large runs on a Mac with 256 GB or more. A published run on M3 Ultra · 256 GB used the Rapid-MLX 4-bit build (source, 2026-09-05).

Smallest Mac with a published run
256 GBM3 Ultra · 256 GB · Rapid-MLX 4-bit
Smallest complete download
16.36 GBQ4_0 GGUF
Files read
2026-10-24Hugging Face

Stability AI Community License. Commercial use: restricted. Excluded territories: none listed. Commercial use requires registration with Stability AI. Licenses terminate above USD $1,000,000 in annual revenue until a separate license is requested.

See the exact GGUF files and sizes

This is Stable Diffusion 3.5 Large, not Large Turbo or Medium.

Table 1

What do you download?

Only builds with every required component and an exact probed file size are shown.

Rapid-MLX 4-bit

mlx · 4-bit · 16.54 GB complete · Recommended

Mac routes: Rapid-MLX

Exact component files for Rapid-MLX 4-bit
FileComponentRequiredDownloadSource
sd3.5_large_4bit_quantized.safetensorsargmaxinc/mlx-stable-diffusion-3.5-large-4bit-quantizedMM-DiT denoiserditYes4.95 GBHugging FaceRead 2026-10-24
text_encoders/clip_l.safetensorsstabilityai/stable-diffusion-3.5-largeCLIP-L text encodertext_encoderYes0.25 GBHugging FaceRead 2026-10-24
text_encoders/clip_g.safetensorsstabilityai/stable-diffusion-3.5-largeOpenCLIP bigG text encodertext_encoderYes1.39 GBHugging FaceRead 2026-10-24
text_encoders/t5xxl_fp16.safetensorsstabilityai/stable-diffusion-3.5-largeT5-XXL text encodertext_encoderYes9.79 GBHugging FaceRead 2026-10-24
vae/diffusion_pytorch_model.safetensorsstabilityai/stable-diffusion-3.5-largeVAEvaeYes0.17 GBHugging FaceRead 2026-10-24

Build sources: Rapid-MLX SD 3.5 Large dogfood · DiffusionKit 4-bit checkpoint card.

Q4_0 GGUF

gguf · Q4_0 · 16.36 GB complete

Mac routes: ComfyUI

Exact component files for Q4_0 GGUF
FileComponentRequiredDownloadSource
sd3.5_large-Q4_0.ggufcity96/stable-diffusion-3.5-large-ggufMM-DiT denoiserditYes4.77 GBHugging FaceRead 2026-10-24
text_encoders/clip_l.safetensorsstabilityai/stable-diffusion-3.5-largeCLIP-L text encodertext_encoderYes0.25 GBHugging FaceRead 2026-10-24
text_encoders/clip_g.safetensorsstabilityai/stable-diffusion-3.5-largeOpenCLIP bigG text encodertext_encoderYes1.39 GBHugging FaceRead 2026-10-24
text_encoders/t5xxl_fp16.safetensorsstabilityai/stable-diffusion-3.5-largeT5-XXL text encodertext_encoderYes9.79 GBHugging FaceRead 2026-10-24
vae/diffusion_pytorch_model.safetensorsstabilityai/stable-diffusion-3.5-largeVAEvaeYes0.17 GBHugging FaceRead 2026-10-24

Build sources: SD 3.5 Large GGUF card · ComfyUI-GGUF README.

Q8_0 GGUF

gguf · Q8_0 · 20.37 GB complete

Mac routes: ComfyUI

Exact component files for Q8_0 GGUF
FileComponentRequiredDownloadSource
sd3.5_large-Q8_0.ggufcity96/stable-diffusion-3.5-large-ggufMM-DiT denoiserditYes8.78 GBHugging FaceRead 2026-10-24
text_encoders/clip_l.safetensorsstabilityai/stable-diffusion-3.5-largeCLIP-L text encodertext_encoderYes0.25 GBHugging FaceRead 2026-10-24
text_encoders/clip_g.safetensorsstabilityai/stable-diffusion-3.5-largeOpenCLIP bigG text encodertext_encoderYes1.39 GBHugging FaceRead 2026-10-24
text_encoders/t5xxl_fp16.safetensorsstabilityai/stable-diffusion-3.5-largeT5-XXL text encodertext_encoderYes9.79 GBHugging FaceRead 2026-10-24
vae/diffusion_pytorch_model.safetensorsstabilityai/stable-diffusion-3.5-largeVAEvaeYes0.17 GBHugging FaceRead 2026-10-24

Build sources: SD 3.5 Large GGUF card · ComfyUI-GGUF README.

Original safetensors

safetensors · Original weights · 28.05 GB complete

Mac routes: ComfyUI

Exact component files for Original safetensors
FileComponentRequiredDownloadSource
sd3.5_large.safetensorsstabilityai/stable-diffusion-3.5-largeMM-DiT denoiserditYes16.46 GBHugging FaceRead 2026-10-24
text_encoders/clip_l.safetensorsstabilityai/stable-diffusion-3.5-largeCLIP-L text encodertext_encoderYes0.25 GBHugging FaceRead 2026-10-24
text_encoders/clip_g.safetensorsstabilityai/stable-diffusion-3.5-largeOpenCLIP bigG text encodertext_encoderYes1.39 GBHugging FaceRead 2026-10-24
text_encoders/t5xxl_fp16.safetensorsstabilityai/stable-diffusion-3.5-largeT5-XXL text encodertext_encoderYes9.79 GBHugging FaceRead 2026-10-24
vae/diffusion_pytorch_model.safetensorsstabilityai/stable-diffusion-3.5-largeVAEvaeYes0.17 GBHugging FaceRead 2026-10-24

Build sources: SD 3.5 Large model card.

Table 2

Does it fit your Mac's RAM?

Published names the exact measured machine. est. is a memory calculation. unknown means the run verdict is not established, not that the model fails.

Run verdicts by Mac memory tier
Mac RAMComplete buildDownloadRun verdictBasis
8 GBRapid-MLX 4-bit4-bit · Rapid-MLX16.54 GBDoes not fitest.
16 GBRapid-MLX 4-bit4-bit · Rapid-MLX16.54 GBDoes not fitest.
24 GBRapid-MLX 4-bit4-bit · Rapid-MLX16.54 GBunknownunknown
32 GBRapid-MLX 4-bit4-bit · Rapid-MLX16.54 GBunknownunknown
48 GBRapid-MLX 4-bit4-bit · Rapid-MLX16.54 GBunknownunknown
64 GBRapid-MLX 4-bit4-bit · Rapid-MLX16.54 GBunknownunknown
96 GBRapid-MLX 4-bit4-bit · Rapid-MLX16.54 GBunknownunknown
128 GBRapid-MLX 4-bit4-bit · Rapid-MLX16.54 GBunknownunknown

How does it rank in blind votes?

#81 of 82 in text to image blind votes, from 23,396 votes.

Quality: blind human votes from stable-diffusion-v35-large (arena.ai), CC BY 4.0, snapshot 2026-10-07. Scores are for the Arena checkpoint, not your local quantized build.

Table 3

What published Mac times are available?

Third-party results keep their original machine and settings. ModelFit does not run these benchmarks.

Published Mac generation measurements
Machine and RAMSettingsAppResultSource
Apple M3 Ultra · 256 GB RAM512x512 · 28 steps · 4-bit · 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.Rapid-MLXSuccess57.3 s16,667,122,480 bytesRapid-MLX SD 3.5 Large dogfoodPublished 2026-09-05 · Grade A

Which Mac software routes are listed?

  • ComfyUI

    Official

    Formats: safetensors, gguf. Memory policy: component stages stay resident.

    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: ComfyUI README · ComfyUI system requirements · ComfyUI CLI arguments · ComfyUI-GGUF README

  • Diffusers

    Official

    Formats: safetensors, gguf. Memory policy: component stages stay resident.

    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: Diffusers MPS guide · Diffusers GGUF guide

  • Rapid-MLX

    Community

    Formats: safetensors, mlx. Memory policy: component stages stay resident.

    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: Rapid-MLX SDXL Base dogfood · Rapid-MLX SD 3.5 Large dogfood

  • stable-diffusion.cpp

    Official

    Formats: gguf, safetensors. Memory policy: component stages stay resident.

    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: stable-diffusion.cpp README, backends

What does the official configuration say?

"The launch page says the model suits professional use at 1 megapixel resolution. It gives no Large-specific RAM or VRAM minimum."

Source: Stable Diffusion 3.5 announcement · Read 2026-10-24.

What known issues should you check?

  • Known issue 1

    One reported ComfyUI run of SD 3.5 Large on a 16 GB Mac mini M4 failed with an MPS memory error. This is one issue report, not a general 16 GB verdict.

    Source: ComfyUI issue 7171 · Read 2026-10-24

Frequently asked questions

How fast is SD 3.5 Large on a Mac?

Rapid-MLX with the 4-bit MLX checkpoint made a 512x512 image in 28 steps in 57.29 s on an M3 Ultra with 256 GiB. The OS peak memory footprint was 16,667,122,480 bytes. The first run took 62.17 s.

Is SD 3.5 Large free for commercial use?

Only with registration. The Stability AI Community License requires commercial users of this 8.1 billion parameter model to register with Stability AI, and licenses terminate above USD $1,000,000 in annual revenue until a separate license is requested.

Will SD 3.5 Large run on a 16 GB Mac?

No verdict exists. One ComfyUI report on a 16 GB Mac mini M4 failed with an MPS memory error, but it is a single issue report, not a general 16 GB result. The official launch page gives no RAM minimum.

Sources