Files read on Hugging Face: 2026-10-31

Z-Image-Turbo on Mac: files, RAM and published times

Turbo checkpoint of the 6B Z-Image family, released under Apache 2.0. The card example makes a 1024 x 1024 image in 9 scheduler steps, which run 8 DiT forwards, with guidance 0.

Answer

Z-Image-Turbo runs on a Mac with 64 GB or more. A published run on M1 Max · 64 GB used the mflux 4-bit build (source, 2026-10-03).

Smallest Mac with a published run
64 GBM1 Max · 64 GB · mflux 4-bit
Smallest complete download
12.06 GBQ4_K_M GGUF
Files read
2026-10-31Hugging Face

Apache 2.0. Commercial use: allowed. Excluded territories: none listed.

See the exact GGUF files and sizes

Table 1

What do you download?

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

mflux 4-bit

safetensors · 4-bit (mflux --quantize 4) · 32.83 GB complete · Recommended

Mac routes: mflux

Exact component files for mflux 4-bit
FileComponentRequiredDownloadSource
transformer/diffusion_pytorch_model-00001-of-00003.safetensorsTongyi-MAI/Z-Image-TurboZ-Image transformerditYes9.97 GBHugging FaceRead 2026-10-31
transformer/diffusion_pytorch_model-00002-of-00003.safetensorsTongyi-MAI/Z-Image-TurboZ-Image transformerditYes9.97 GBHugging FaceRead 2026-10-31
transformer/diffusion_pytorch_model-00003-of-00003.safetensorsTongyi-MAI/Z-Image-TurboZ-Image transformerditYes4.67 GBHugging FaceRead 2026-10-31
text_encoder/model-00001-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes3.96 GBHugging FaceRead 2026-10-31
text_encoder/model-00002-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes3.99 GBHugging FaceRead 2026-10-31
text_encoder/model-00003-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes0.10 GBHugging FaceRead 2026-10-31
vae/diffusion_pytorch_model.safetensorsTongyi-MAI/Z-Image-TurboVAEvaeYes0.17 GBHugging FaceRead 2026-10-31

Build sources: mflux model registry · mflux README.

Q4_K_M GGUF

gguf · Q4_K_M · 12.06 GB complete

Mac routes: stable-diffusion.cpp

Exact component files for Q4_K_M GGUF
FileComponentRequiredDownloadSource
z-image-turbo-Q4_K_M.ggufunsloth/Z-Image-Turbo-GGUFZ-Image transformerditYes3.85 GBHugging FaceRead 2026-10-31
text_encoder/model-00001-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes3.96 GBHugging FaceRead 2026-10-31
text_encoder/model-00002-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes3.99 GBHugging FaceRead 2026-10-31
text_encoder/model-00003-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes0.10 GBHugging FaceRead 2026-10-31
vae/diffusion_pytorch_model.safetensorsTongyi-MAI/Z-Image-TurboVAEvaeYes0.17 GBHugging FaceRead 2026-10-31

Build sources: Z-Image-Turbo GGUF card · stable-diffusion.cpp README.

Q8_0 GGUF

gguf · Q8_0 · 15.44 GB complete

Mac routes: stable-diffusion.cpp

Exact component files for Q8_0 GGUF
FileComponentRequiredDownloadSource
z-image-turbo-Q8_0.ggufunsloth/Z-Image-Turbo-GGUFZ-Image transformerditYes7.22 GBHugging FaceRead 2026-10-31
text_encoder/model-00001-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes3.96 GBHugging FaceRead 2026-10-31
text_encoder/model-00002-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes3.99 GBHugging FaceRead 2026-10-31
text_encoder/model-00003-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes0.10 GBHugging FaceRead 2026-10-31
vae/diffusion_pytorch_model.safetensorsTongyi-MAI/Z-Image-TurboVAEvaeYes0.17 GBHugging FaceRead 2026-10-31

Build sources: Z-Image-Turbo GGUF card · stable-diffusion.cpp README.

Original safetensors

safetensors · Original weights · 32.83 GB complete

Mac routes: Diffusers

Exact component files for Original safetensors
FileComponentRequiredDownloadSource
transformer/diffusion_pytorch_model-00001-of-00003.safetensorsTongyi-MAI/Z-Image-TurboZ-Image transformerditYes9.97 GBHugging FaceRead 2026-10-31
transformer/diffusion_pytorch_model-00002-of-00003.safetensorsTongyi-MAI/Z-Image-TurboZ-Image transformerditYes9.97 GBHugging FaceRead 2026-10-31
transformer/diffusion_pytorch_model-00003-of-00003.safetensorsTongyi-MAI/Z-Image-TurboZ-Image transformerditYes4.67 GBHugging FaceRead 2026-10-31
text_encoder/model-00001-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes3.96 GBHugging FaceRead 2026-10-31
text_encoder/model-00002-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes3.99 GBHugging FaceRead 2026-10-31
text_encoder/model-00003-of-00003.safetensorsTongyi-MAI/Z-Image-TurboQwen3 text encodertext_encoderYes0.10 GBHugging FaceRead 2026-10-31
vae/diffusion_pytorch_model.safetensorsTongyi-MAI/Z-Image-TurboVAEvaeYes0.17 GBHugging FaceRead 2026-10-31

Build sources: Z-Image-Turbo 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 GBQ4_K_M GGUFQ4_K_M · stable-diffusion.cpp12.06 GBDoes not fitest.
16 GBQ4_K_M GGUFQ4_K_M · stable-diffusion.cpp12.06 GBunknownunknown
24 GBQ4_K_M GGUFQ4_K_M · stable-diffusion.cpp12.06 GBunknownunknown
32 GBQ4_K_M GGUFQ4_K_M · stable-diffusion.cpp12.06 GBunknownunknown
48 GBmflux 4-bit4-bit (mflux --quantize 4) · mflux32.83 GBunknownunknown
64 GBmflux 4-bit4-bit (mflux --quantize 4) · mflux32.83 GBRuns · PublishedPublishedApple M1 Max · 64 GB RAM
96 GBmflux 4-bit4-bit (mflux --quantize 4) · mflux32.83 GBRuns · Published on M1 Max · 64 GBPublishedApple M1 Max · 64 GB RAM
128 GBmflux 4-bit4-bit (mflux --quantize 4) · mflux32.83 GBRuns · Published on M1 Max · 64 GBPublishedApple M1 Max · 64 GB RAM

Every image model, ranked by memory: 8 GB Mac · 16 GB Mac · 24 GB Mac · 32 GB Mac · 64 GB Mac

Table 3

Does it fit your NVIDIA GPU's VRAM?

Runs · Stated means the linked model card, app maker or NVIDIA says a GPU with that much VRAM runs the model; Runs · Published names the card of a dated run. Fits with offload means the main model weights fit in VRAM with a 15% margin but the full set does not: the text encoder and VAE run from system RAM, which is slower. Exceeds VRAM means even the main weights are too large for the card. Mac runs never count here.

Run verdicts by NVIDIA VRAM tier
NVIDIA VRAMComplete buildDownloadRun verdictBasis
8 GBNot build-specificunknownRuns · StatedStated4 GB · stable-diffusion.cpp with GGUF weights · stable-diffusion.cpp Z-Image docs
12 GBNot build-specificunknownRuns · StatedStated4 GB · stable-diffusion.cpp with GGUF weights · stable-diffusion.cpp Z-Image docs
16 GBNot build-specificunknownRuns · StatedStated16 GB · stated for consumer GPUs · Z-Image-Turbo model card
24 GBNot build-specificunknownRuns · StatedStated16 GB · stated for consumer GPUs · Z-Image-Turbo model card
32 GBNot build-specificunknownRuns · StatedStated16 GB · stated for consumer GPUs · Z-Image-Turbo model card

Every image model, ranked by memory: 8 GB of VRAM · 12 GB of VRAM · 16 GB of VRAM · 24 GB of VRAM

How does it rank in blind votes?

#59 of 82 in text to image blind votes, from 24,808 votes.

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

Table 4

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 M1 Max · 64 GB RAM1024x1024 · 8 steps · 4-bit (mflux --quantize 4) · 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.mfluxSuccess102.4 s26.1 GB peakmflux pull request 820Published 2026-10-03 · Grade A

Which Mac software routes are listed?

  • 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

  • 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 card says the model fits comfortably within 16G VRAM consumer devices. This is a GPU statement for the Z-Image family, not a Mac minimum."

Source: Z-Image-Turbo model card · Read 2026-10-24.

What known issues should you check?

  • Known issue 1

    The only Mac run is a pre-release test in an mflux pull request, made before mflux changed its default float32 stream. Treat it as one data point.

    Source: mflux pull request 820 · Read 2026-10-24

Frequently asked questions

Can I run Z-Image-Turbo on a Mac?

One published Mac run exists. mflux with 4-bit quantization made a 1024x1024 image in 8 steps on an M1 Max with 64 GB in 102.4 s, peaking at 26.13 GB of MLX memory. It predates a later mflux change, so treat it as one data point, not a minimum.

Is Z-Image-Turbo free for commercial use?

Yes. Z-Image-Turbo is released under Apache 2.0, which allows commercial use. The card describes the 6B Z-Image family and says it fits comfortably within 16G VRAM consumer devices, a GPU statement rather than a Mac requirement.

Which Z-Image-Turbo build should I download?

The measured Mac path is mflux, which loads the original weights and quantizes them to 4-bit while loading. The Q8_0 and Q4_K_M GGUF files cover the transformer only, so they need the text encoder and VAE from the original repository.

Sources