31 PyTorch models converted to Core ML (.mlpackage) for on-device inference on iPhone, iPad and Mac — detection, segmentation, depth, matting, diffusion, TTS, source separation and more. 6.96 GB of packages in one repo.

Every model here has a matching, readable conversion script and — for most of them — a standalone SwiftUI sample app in the CoreML-Models repo.

This repo is also the backing store for the Models Zoo app on the App Store, which downloads and runs any of these models on device without writing code. models.json is that app's manifest — treat it as machine-owned.

Each model also has its own repo with a model card, a demo image and the unzipped .mlpackage — the Model column links there. The In this repo column is the zipped copy the app downloads.

Models

Model	Task	Packages	Size	License	Upstream	In this repo
Face Parsing	image segmentation	1	47 MB	MIT	zllrunning/face-parsing.PyTorch	zip
MobileSAM	mask generation	1	19 MB	Apache-2.0	ChaoningZhang/MobileSAM	zip
RMBG-1.4	image segmentation	1	37 MB	Bria RMBG-1.4 License	briaai/RMBG-1.4	zip
DDColor Tiny	image to image	1	203 MB	Apache-2.0	piddnad/DDColor	zip
Pixelization	image to image	1	35 MB	Research use only	WuZongWei6/Pixelization	zip
Real-ESRGAN 4x	image to image	1	59 MB	BSD-3-Clause	xinntao/Real-ESRGAN	zip
SinSR	image to image	3	517 MB	CC BY-NC-SA 4.0	wyf0912/SinSR	zip
RF-DETR Nano	object detection	1	95 MB	Apache-2.0	roboflow/rf-detr	zip
YOLO-World	zero shot object detection	2	134 MB	GPL-3.0	AILab-CVC/YOLO-World	zip
YOLO11s	object detection	1	17 MB	AGPL-3.0	ultralytics/ultralytics	zip
YOLO26s	object detection	1	17 MB	AGPL-3.0	ultralytics/ultralytics	zip
YOLOv10n	object detection	1	4 MB	AGPL-3.0	THU-MIG/yolov10	zip
Depth Anything 3 Base (504×504)	depth estimation	1	173 MB	Apache-2.0	ByteDance-Seed/Depth-Anything-3	zip
Depth Anything 3 Small (504×504)	depth estimation	1	44 MB	Apache-2.0	ByteDance-Seed/Depth-Anything-3	zip
MoGe-2 ViT-B (504×504)	depth estimation	1	184 MB	MIT	microsoft/MoGe	zip
Florence-2	image text to text	3	229 MB	MIT	microsoft/Florence-2	zip
SigLIP	zero shot image classification	2	358 MB	Apache-2.0	google-research/big_vision	zip
3DDFA V2	keypoint detection	1	6 MB	MIT	cleardusk/3DDFA_V2	zip
Hyper-SD (1-Step)	text to image	4	905 MB	OpenRAIL-M	ByteDance/Hyper-SD	zip
Nitro-E (4-Step)	text to image	3	987 MB	MIT (Nitro-E) + Llama 3.2 Community License (text encoder)	amd/Nitro-E	zip
MatAnyone	image segmentation	5	83 MB	S-Lab License 1.0	pq-yang/MatAnyone	zip
HTDemucs	audio to audio	1	75 MB	MIT	adefossez/demucs	zip
OpenVoice V2	audio to audio	2	58 MB	MIT	myshell-ai/OpenVoice	zip
Pyannote Diarization	voice activity detection	1	5 MB	MIT	pyannote/pyannote-audio	zip
Kokoro-82M	text to speech	4	724 MB	Apache-2.0	hexgrad/Kokoro-82M	zip
Stable Audio Open	text to audio	4	1.41 GB	Stability AI Community License	stabilityai/stable-audio-open-small	zip
LaMa	image to image	1	187 MB	Apache-2.0	advimman/lama	zip
GFPGAN	image to image	1	298 MB	Apache-2.0	TencentARC/GFPGAN	zip
AdaFace IR-18 †	face recognition	1	42 MB	MIT	mk-minchul/AdaFace	zip
EfficientAD (MVTec bottle) †	anomaly detection	1	14 MB	MIT	openvinotoolkit/anomalib	zip
YOLOE-S †	zero shot object detection + segmentation	3	133 MB	AGPL-3.0	THU-MIG/yoloe	zip

† Downloadable here, but not yet selectable inside the Models Zoo app — the shipped app build has no UI template for them.

Also in the Models Zoo app (hosted in separate repos)

Model	Repo
Gemma 4 E2B	mlboydaisuke/gemma-4-E2B-coreml
Gemma 4 E4B	mlboydaisuke/gemma-4-E4B-coreml
Qwen3-VL 2B	mlboydaisuke/qwen3-vl-2b-coreml
Qwen3.5 0.8B	mlboydaisuke/qwen3.5-0.8B-CoreML
Qwen3.5 2B	mlboydaisuke/qwen3.5-2B-CoreML

Download

Each model lives in its own directory and ships as a zipped .mlpackage.

bash

hf download mlboydaisuke/coreml-zoo --include "moge2/*" --local-dir ./moge2
unzip './moge2/moge2/*.zip' -d ./moge2

bash

# everything (large)
hf download mlboydaisuke/coreml-zoo --local-dir ./coreml-zoo

Use in Swift

swift

import CoreML

let config = MLModelConfiguration()
config.computeUnits = .cpuAndNeuralEngine   // see each model's card

let compiled = try await MLModel.compileModel(at: mlpackageURL)
let model = try MLModel(contentsOf: compiled, configuration: config)

Two things that bite on real devices, both documented per model in docs/coreml_conversion_notes.md:

Compute units are load-bearing. Several models are converted for a specific backend (FP32 + .cpuOnly where FP16 attention overflows, .cpuOnly where the iOS GPU hits the MPS singleton-slice bug). Switching them is not free.
MLMultiArray.dataPointer is not contiguous on the Neural Engine. ANE pads rows for SIMD alignment — always read through array.strides.

Licenses

Licenses are per model and inherited from upstream. Several are non-commercial (MatAnyone: S-Lab 1.0, SinSR: CC BY-NC-SA 4.0, RMBG-1.4: Bria RMBG license, Pixelization: research use only) and the YOLO models are AGPL-3.0 / GPL-3.0. Check the table above and the upstream repo before shipping anything commercially.

Credits

Conversions by john-rocky (Daisuke Majima). Upstream authors are credited per model above.

More models in this format: Core ML Model Zoo — 46 models, each with the recipe that produced it.

Want a different model on-device? Open a request — free, open weights only; the export and its measured numbers get published publicly.