Files read on Hugging Face: 2026-10-24

FLUX.2 [klein] on Mac: files, RAM and published times

Compact FLUX.2 text-to-image and editing family in 4B and 9B sizes. The 4B model is Apache 2.0; the 9B models use the FLUX Non-Commercial License.

Black Forest Labs · Updated 2026-10-24
Answer

FLUX.2 [klein] runs on a Mac with 96 GB or more. A published run on M2 Ultra · 96 GB used the Klein 4B BF16 build (source, 2026-02-11).

Smallest Mac with a published run
96 GBM2 Ultra · 96 GB · Klein 4B BF16
Smallest complete download
12.51 GBKlein 4B Q8_0 GGUF
Files read
2026-10-24Hugging Face

Apache 2.0 (4B), FLUX Non-Commercial License (9B). Commercial use: restricted. Excluded territories: none listed. 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.

See the exact GGUF files and sizes

Builds here are the distilled 4B and 9B checkpoints. Base 4B, Base 9B and 9B KV are separate checkpoints without their own Mac measurements.

Table 1

What do you download?

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

Klein 4B BF16

safetensors · BF16 · 15.96 GB complete · Recommended

Mac routes: Diffusers, flux-2-swift-mlx

Exact component files for Klein 4B BF16
FileComponentRequiredDownloadSource
flux-2-klein-4b.safetensorsblack-forest-labs/FLUX.2-klein-4BFLUX.2 klein transformerditYes7.75 GBHugging FaceRead 2026-10-24
text_encoder/model-00001-of-00002.safetensorsblack-forest-labs/FLUX.2-klein-4BQwen3 text encodertext_encoderYes4.97 GBHugging FaceRead 2026-10-24
text_encoder/model-00002-of-00002.safetensorsblack-forest-labs/FLUX.2-klein-4BQwen3 text encodertext_encoderYes3.08 GBHugging FaceRead 2026-10-24
vae/diffusion_pytorch_model.safetensorsblack-forest-labs/FLUX.2-klein-4BFLUX.2 VAEvaeYes0.17 GBHugging FaceRead 2026-10-24

Build sources: flux-2-swift-mlx model registry · FLUX.2 klein 4B model card.

Klein 4B Q8_0 GGUF

gguf · Q8_0 · 12.51 GB complete

Mac routes: stable-diffusion.cpp

Exact component files for Klein 4B Q8_0 GGUF
FileComponentRequiredDownloadSource
flux-2-klein-4b-Q8_0.ggufunsloth/FLUX.2-klein-4B-GGUFFLUX.2 klein transformerditYes4.30 GBHugging FaceRead 2026-10-24
text_encoder/model-00001-of-00002.safetensorsblack-forest-labs/FLUX.2-klein-4BQwen3 text encodertext_encoderYes4.97 GBHugging FaceRead 2026-10-24
text_encoder/model-00002-of-00002.safetensorsblack-forest-labs/FLUX.2-klein-4BQwen3 text encodertext_encoderYes3.08 GBHugging FaceRead 2026-10-24
vae/diffusion_pytorch_model.safetensorsblack-forest-labs/FLUX.2-klein-4BFLUX.2 VAEvaeYes0.17 GBHugging FaceRead 2026-10-24

Build sources: FLUX.2 klein GGUF card · stable-diffusion.cpp README.

Klein 9B Q4_K_M GGUF

gguf · Q4_K_M · 22.46 GB complete

Mac routes: stable-diffusion.cpp

Exact component files for Klein 9B Q4_K_M GGUF
FileComponentRequiredDownloadSource
flux-2-klein-9b-Q4_K_M.ggufunsloth/FLUX.2-klein-9B-GGUFFLUX.2 klein transformerditYes5.91 GBHugging FaceRead 2026-10-24
text_encoder/model-00001-of-00004.safetensorsblack-forest-labs/FLUX.2-klein-9BQwen3 text encodertext_encoderYes4.90 GBHugging FaceRead 2026-10-24
text_encoder/model-00002-of-00004.safetensorsblack-forest-labs/FLUX.2-klein-9BQwen3 text encodertext_encoderYes4.92 GBHugging FaceRead 2026-10-24
text_encoder/model-00003-of-00004.safetensorsblack-forest-labs/FLUX.2-klein-9BQwen3 text encodertext_encoderYes4.98 GBHugging FaceRead 2026-10-24
text_encoder/model-00004-of-00004.safetensorsblack-forest-labs/FLUX.2-klein-9BQwen3 text encodertext_encoderYes1.58 GBHugging FaceRead 2026-10-24
vae/diffusion_pytorch_model.safetensorsblack-forest-labs/FLUX.2-klein-9BFLUX.2 VAEvaeYes0.17 GBHugging FaceRead 2026-10-24

Build sources: FLUX.2 klein GGUF card · stable-diffusion.cpp README.

Klein 9B BF16

safetensors · BF16 · 34.71 GB complete

Mac routes: Diffusers, flux-2-swift-mlx

Exact component files for Klein 9B BF16
FileComponentRequiredDownloadSource
flux-2-klein-9b.safetensorsblack-forest-labs/FLUX.2-klein-9BFLUX.2 klein transformerditYes18.16 GBHugging FaceRead 2026-10-24
text_encoder/model-00001-of-00004.safetensorsblack-forest-labs/FLUX.2-klein-9BQwen3 text encodertext_encoderYes4.90 GBHugging FaceRead 2026-10-24
text_encoder/model-00002-of-00004.safetensorsblack-forest-labs/FLUX.2-klein-9BQwen3 text encodertext_encoderYes4.92 GBHugging FaceRead 2026-10-24
text_encoder/model-00003-of-00004.safetensorsblack-forest-labs/FLUX.2-klein-9BQwen3 text encodertext_encoderYes4.98 GBHugging FaceRead 2026-10-24
text_encoder/model-00004-of-00004.safetensorsblack-forest-labs/FLUX.2-klein-9BQwen3 text encodertext_encoderYes1.58 GBHugging FaceRead 2026-10-24
vae/diffusion_pytorch_model.safetensorsblack-forest-labs/FLUX.2-klein-9BFLUX.2 VAEvaeYes0.17 GBHugging FaceRead 2026-10-24

Build sources: flux-2-swift-mlx model registry · FLUX.2 klein 9B 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 GBKlein 4B BF16BF16 · flux-2-swift-mlx15.96 GBDoes not fitest.
16 GBKlein 4B BF16BF16 · flux-2-swift-mlx15.96 GBunknownunknown
24 GBKlein 4B BF16BF16 · flux-2-swift-mlx15.96 GBunknownunknown
32 GBKlein 4B BF16BF16 · flux-2-swift-mlx15.96 GBunknownunknown
48 GBKlein 4B BF16BF16 · flux-2-swift-mlx15.96 GBunknownunknown
64 GBKlein 4B BF16BF16 · flux-2-swift-mlx15.96 GBunknownunknown
96 GBKlein 4B BF16BF16 · flux-2-swift-mlx15.96 GBRuns · PublishedPublishedApple M2 Ultra · 96 GB RAM
128 GBKlein 4B BF16BF16 · flux-2-swift-mlx15.96 GBRuns · Published on M2 Ultra · 96 GBPublishedApple M2 Ultra · 96 GB RAM

How does it rank in blind votes?

No independent score yet.

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 M2 Ultra · 96 GB RAM1024x1024 · 4 steps · BF16 · 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.flux-2-swift-mlxSuccess25.6 s17 846 MBflux-2-swift-mlx quantization benchmarkPublished 2026-02-11 · Grade A
Apple M2 Ultra · 96 GB RAM1024x1024 · 4 steps · BF16 · 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.flux-2-swift-mlxSuccess54.9 s27 779 MBflux-2-swift-mlx quantization benchmarkPublished 2026-02-11 · 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

  • Draw Things

    Official

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

    Native offline app for Mac. It picks a weight-streaming tier from physical memory. No official Mac timing or peak-memory table was found, so there is no run verdict here.

    Sources: Draw Things on the App Store · Draw Things App Store listing data · Draw Things model importer

  • flux-2-swift-mlx

    Community

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

    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: flux-2-swift-mlx README · flux-2-swift-mlx model registry

  • mflux

    Community · 0.23.0

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

    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.

    Install with uv

    uv tool install --upgrade mflux

    Sources: mflux README, install · mflux pyproject · mflux common README, quantization

  • 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 4B card says the model fits in about 13GB VRAM and is accessible on NVIDIA RTX 3090/4070 and above. This is an NVIDIA statement, not a Mac minimum."

Source: FLUX.2 klein 4B model card · Read 2026-10-24.

What known issues should you check?

  • Known issue 1

    Official sources disagree on the 4B VRAM figure: the Hugging Face card says about 13GB, while the reference README says about 8GB without naming a configuration.

    Source: FLUX.2 reference README · Read 2026-10-24

  • Known issue 2

    The 9B card says the model fits in about 29GB VRAM and is accessible on NVIDIA RTX 4090 and above. This is an NVIDIA statement, not a Mac minimum.

    Source: FLUX.2 klein 9B model card · Read 2026-10-24

Frequently asked questions

How fast is FLUX.2 klein on a Mac?

On an M2 Ultra with 96 GB, flux-2-swift-mlx made a 1024x1024 image in 4 steps in 25.6 s for Klein 4B and 54.9 s for Klein 9B at BF16. Peaks were 17.846 GB and 27.779 GB, covering transformer, VAE and activations only.

Is FLUX.2 klein free for commercial use?

The 4B model is Apache 2.0 and allows commercial use. The 9B models use the FLUX Non-Commercial License, which limits model use to non-commercial work, and filters or manual review are required.

How much VRAM does Klein 4B need?

The Hugging Face card says Klein 4B fits in about 13GB VRAM on NVIDIA RTX 3090/4070 and above, while the reference README says about 8GB without naming a configuration. Neither figure is a Mac minimum.

Sources