Skip to content

All on-device models

Image generation models for iPhone and Mac (Core ML & MLX)

102 models for this task are tagged Core ML or MLX on Hugging Face, or come from Apple: 15 Core ML, 86 MLX and 1 other Apple models. The table lists the most downloaded in the last 30 days.

Last updated 2026-11-02 (UTC).

Showing 100 of 100.

ModelSizeDownloads (30 days)
Runpod/FLUX.2-klein-4B-mflux-4bitMLXfrom black-forest-labs/FLUX.2-klein-4B4.6 GB68.5K
mflux-community/z-image-turbo-mflux-q4MLXfrom Tongyi-MAI/Z-Image-Turbo5.9 GB29.6K
mflux-community/z-image-turbo-mflux-q8MLXfrom Tongyi-MAI/Z-Image-Turbo11 GB25.4K
mlx-community/Qwen-Image-2.1-mflux-q4MLXfrom Qwen/Qwen-Image-2.19.6 GB22.1K
mflux-community/qwen-image-2-1-turbo-mflux-q4MLXfrom Qwen/Qwen-Image-2.1-Turbo10 GB16.4K
mlx-community/Qwen-Image-2512-4bitMLXfrom Qwen/Qwen-Image-251225.9 GB15.7K
mflux-community/qwen-image-2-1-turbo-mflux-q8MLXfrom Qwen/Qwen-Image-2.1-Turbo17.8 GB14.6K
mflux-community/flux-1-schnell-mflux-q4MLXfrom black-forest-labs/FLUX.1-schnell9.6 GB14.1K
abenzerps/Qwen-Image-2.1-Turbo-QuantizedMLXfrom Qwen/Qwen-Image-2.1-Turbo90 GB11.7K
mflux-community/qwen-image-2-1-mflux-q4MLXfrom Qwen/Qwen-Image-2.110.6 GB10K
mflux-community/krea-2-turbo-mflux-q8MLXfrom krea/Krea-2-Turbo22.2 GB9,697
mflux-community/qwen-image-2512-mflux-q8MLXfrom Qwen/Qwen-Image-251236.1 GB9,370
mflux-community/krea-2-turbo-mflux-q4MLXfrom krea/Krea-2-Turbo15.8 GB7,514
JoyFusionAI/Qwen-Image-2.1-MLX-4bitMLXfrom Qwen/Qwen-Image-2.120.5 GB7,467
mflux-community/flux-3-image-klein-mflux-q8MLXfrom black-forest-labs/flux-3-image-klein9.9 GB7,242
mflux-community/qwen-image-2512-mflux-q4MLXfrom Qwen/Qwen-Image-2512—5,604
mlx-community/Qwen-Image-2512-8bitMLXfrom Qwen/Qwen-Image—5,202
mflux-community/qwen-image-2-1-controlnet-union-mflux-q4MLXfrom Qwen/Qwen-Image-2.1—4,969
mflux-community/z-image-turbo-mflux-bf16MLXfrom Tongyi-MAI/Z-Image-Turbo—4,251
mflux-community/flux-3-image-klein-mflux-bf16MLXfrom black-forest-labs/flux-3-image-klein—3,780
mflux-community/flux-1-dev-fill-mflux-q4MLXfrom black-forest-labs/FLUX.1-Fill-dev—3,495
argmaxinc/mlx-FLUX.1-schnell-4bit-quantizedMLX—3,441
JoyFusionAI/Qwen-Image-2.1-MLX-8bitMLXfrom Qwen/Qwen-Image-2.1—3,303
mflux-community/qwen-image-2-1-mflux-bf16MLXfrom Qwen/Qwen-Image-2.1—3,193
mflux-community/flux-3-image-klein-mflux-q4MLXfrom black-forest-labs/flux-3-image-klein—3,064
mflux-community/qwen-image-2-1-mflux-q8MLXfrom Qwen/Qwen-Image-2.1—3,060
mflux-community/flux-1-schnell-mflux-q8MLXfrom black-forest-labs/FLUX.1-schnell—2,458
joeynyc/Studio-Image-Turbo-MLX-6bit-from-Qwen-2.1MLXfrom Qwen/Qwen-Image-2.1-Turbo—2,271
ddalcu/Qwen-Image-2.1-MLX-Serve-8bitMLXfrom Qwen/Qwen-Image-2.1—1,994
MLXBits/krea-2-mlx-q8MLXfrom krea/Krea-2-Turbo—1,937
mlx-community/Qwen-Image-2512-6bitMLXfrom Qwen/Qwen-Image-2512—1,909
mflux-community/qwen-image-2-1-turbo-mflux-q6MLXfrom Qwen/Qwen-Image-2.1-Turbo—1,821
mflux-community/moss-image-sft-mflux-q4MLXfrom OpenMOSS-Team/MOSS-Image-SFT—1,796
joeynyc/Studio-Image-Turbo-MLX-4bit-from-Qwen-2.1MLXfrom Qwen/Qwen-Image-2.1-Turbo—1,728
mflux-community/moss-image-sft-mflux-q8MLXfrom OpenMOSS-Team/MOSS-Image-SFT—1,718
mflux-community/qwen-image-mflux-bf16MLXfrom Qwen/Qwen-Image—1,703
mh73772/krea-2-turbo-mflux-q8-4stepMLXfrom krea/Krea-2-Turbo—1,680
mflux-community/z-image-turbo-mflux-q6MLXfrom Tongyi-MAI/Z-Image-Turbo—1,666
vanch007/FLUX.3-Image-Klein-mlx-4bitMLX—1,433
mflux-community/flux-1-dev-fill-mflux-q8MLXfrom black-forest-labs/FLUX.1-Fill-dev—1,418
mflux-community/ming-image-0-1-mflux-q3MLXfrom inclusionAI/Ming-Image-0.1-Design—1,359
joeynyc/Ming-Image-0.1-Design-mflux-q8-te5MLXfrom inclusionAI/Ming-Image-0.1-Design—1,314
mflux-community/qwen-image-2512-mflux-q6MLXfrom Qwen/Qwen-Image-2512—1,241
NovaeonStudio/qwen-image-2.1-mflux-q8MLXfrom Qwen/Qwen-Image-2.1—1,199
vanch007/FLUX.3-Image-Klein-mlx-bf16MLX—1,161
ddalcu/Qwen-Image-2.1-MLX-Serve-4bitMLXfrom Qwen/Qwen-Image-2.1—1,159
SirSahOl/Qwen-Image-mlx-16bitMLXfrom Qwen/Qwen-Image—1,156
mflux-community/fibo-2-0-edit-turbo-mflux-q4MLXfrom briaai/Fibo-2.0-edit-turbo—1,156
mflux-community/krea-2-raw-mflux-q4MLXfrom krea/Krea-2-Raw—1,123
mflux-community/flux-3-image-klein-mflux-q6MLXfrom black-forest-labs/flux-3-image-klein—1,121
mflux-community/qwen-image-mflux-q4MLXfrom Qwen/Qwen-Image—1,100
deepsweet/Z-Image-Turbo-6B-MLX-Q8MLXfrom Tongyi-MAI/Z-Image-Turbo—1,099
mflux-community/qwen-image-mflux-q6MLXfrom Qwen/Qwen-Image—1,054
mflux-community/fibo-lite-mflux-q4MLXfrom briaai/Fibo-lite—1,025
mflux-community/ernie-image-base-mflux-q4MLXfrom baidu/ERNIE-Image—1,002
MLXBits/krea-2-mlx-q4MLXfrom krea/Krea-2-Turbo—991
ddalcu/Qwen-Image-2.1-Turbo-MLX-Serve-8bitMLXfrom Qwen/Qwen-Image-2.1-Turbo—945
argmaxinc/mlx-FLUX.1-schnellMLX—924
mflux-community/qwen-image-2512-mflux-bf16MLXfrom Qwen/Qwen-Image-2512—920
mflux-community/z-image-turbo-mflux-q5MLXfrom Tongyi-MAI/Z-Image-Turbo—906
prism-ml/bonsai-image-ternary-4B-mlx-2bitMLXfrom prism-ml/bonsai-image-ternary-4B-unpacked—893
mflux-community/flux-1-dev-mflux-q8MLXfrom black-forest-labs/FLUX.1-dev—889
mflux-community/ernie-image-turbo-mflux-q8MLXfrom baidu/ERNIE-Image-Turbo—887
argmaxinc/mlx-stable-diffusion-3.5-largeMLXfrom stabilityai/stable-diffusion-3.5-large—884
mflux-community/qwen-image-mflux-q8MLXfrom Qwen/Qwen-Image—870
mflux-community/ideogram-4-mflux-q3MLXfrom ideogram-ai/ideogram-4-fp8—863
mflux-community/z-image-base-mflux-q8MLXfrom Tongyi-MAI/Z-Image—853
vanch007/FLUX.3-Image-Klein-mlx-8bitMLX—816
joeynyc/Ming-Image-0.1-Design-mflux-q8-te8MLXfrom inclusionAI/Ming-Image-0.1-Design—805
mflux-community/fibo-2-0-dev-mflux-q8MLXfrom briaai/Fibo-2.0-dev—798
mflux-community/qwen-image-2-1-turbo-mflux-bf16MLXfrom Qwen/Qwen-Image-2.1-Turbo—795
mflux-community/flux-1-dev-controlnet-canny-mflux-q8MLXfrom black-forest-labs/FLUX.1-dev—778
mflux-community/qwen-image-2-1-mflux-q3MLXfrom Qwen/Qwen-Image-2.1—767
SirSahOl/Qwen-Image-mlx-4bitMLXfrom Qwen/Qwen-Image—752
mflux-community/flux-1-dev-mflux-q4MLXfrom black-forest-labs/FLUX.1-dev—751
deepsweet/Z-Image-Turbo-6B-MLX-Q4MLXfrom Tongyi-MAI/Z-Image-Turbo—693
mflux-community/iris-3b-turbo-mflux-q4MLXfrom speridlabs/iris-3b—662
mlx-community/Qwen-Image-2512-3bitMLXfrom Qwen/Qwen-Image-2512—660
mflux-community/ernie-image-turbo-mflux-q4MLXfrom baidu/ERNIE-Image-Turbo—646
mflux-community/krea-2-turbo-mflux-bf16MLXfrom krea/Krea-2-Turbo—642
ddalcu/Krea-2-Turbo-MLX-Serve-mixed-4-8MLXfrom krea/Krea-2-Turbo—627
ddalcu/Qwen-Image-2.1-Turbo-MLX-Serve-4bitMLXfrom Qwen/Qwen-Image-2.1-Turbo—622
mflux-community/flux-1-dev-fill-mflux-q6MLXfrom black-forest-labs/FLUX.1-Fill-dev—611
mflux-community/qwen-image-2-1-mflux-q6MLXfrom Qwen/Qwen-Image-2.1—609
SirSahOl/Qwen-Image-mlx-8bitMLXfrom Qwen/Qwen-Image—603
argmaxinc/mlx-stable-diffusion-3.5-large-4bit-quantizedMLXfrom stabilityai/stable-diffusion-3.5-large—580
apple/coreml-stable-diffusion-v1-5Core ML—63
apple/coreml-stable-diffusion-2-1-baseCore ML—47
devin-lai/Qwen-Image-2.1-CoremlCore MLfrom Qwen/Qwen-Image-2.1—38
mlboydaisuke/Hyper-SD-1step-CoreMLCore MLfrom ByteDance/Hyper-SD—37
digitalbrain79/sdxl-lightning-4step-coreml-6bits-compiledCore ML—29
c2p-cmd/StableDiffusionXLBase1.0-Palletized6.5Core ML—22
apple/coreml-stable-diffusion-2-1-base-palettizedCore ML—20
darkmaniac7/tokforge-ios-coreml-image-modelsCore MLfrom stabilityai/sdxl-turbo—20
apple/coreml-stable-diffusion-xl-baseCore ML—19
apple/coreml-stable-diffusion-v1-4Core ML—17
apple/coreml-stable-diffusion-2-baseCore ML—11
apple/coreml-stable-diffusion-v1-5-palettizedCore ML—11
apple/coreml-stable-diffusion-xl-base-with-refinerCore ML—11
apple/coreml-stable-diffusion-2-base-palettizedCore ML—9

Reading the table

  • Downloads are Hugging Face's count for the last 30 days, not all time. Hugging Face counts requests for a model's main file, so one install can count more than once.
  • Size is the total of the repo's current files on Hugging Face, in the sizes it shows (1 GB = 1,000 MB). Some repos hold several variants of a model, so it can be larger than the one download you need; a dash means it has not been measured yet.
  • Trending is Hugging Face's own score for recent attention. "from" under a name is the model it was made from.
  • Parameters is the parameter count in the model files. A count marked * comes from the model's name (such as 27B) instead: the files gave none, or a much smaller one, as packed low-bit files can.
  • Quantization is read from the model's tags and name, not checked against the files.

Before you use one

The list comes from Hugging Face's public API and is updated daily. We don't test each model. Check the license and read the model card before you ship one in an app: some licenses limit commercial use, and "other" means the terms are on the model card. Models whose names or tags point at adult or safety-removed content are left out, and so are gated models that need a Hugging Face sign-in to download.