Unable to load Stable Diffusion v1.5 CoreML model

[cordschneider]

Thank you for the incredible work that you have put into creating this repo!

The installation worked extremely smoothly, and using the suggested CyberRealistic_V8_FP16.safetensors model generated a test image perfectly. As I'm on a Mac, I wanted to try the Stable Diffusion v1.5 CoreML (split_einsum, palettized 6-bit) model:

Host Specifications

• CPU: Apple M1 Pro
• GPU: Apple M1 Pro
• Memory: 32 GB RAM (5 Cores)
• OS: Darwin 25.5.0

The model downloaded correctly, but selecting Load resulted in Model Load Failed. exited with code 2 message being displayed. Looking through the logs, I can see the following:

  [download] Starting download of coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled.zip from https://cas-bridge.xethub.hf.co/xet-bridge-us/648996fd195a6f09f62fb4be/569fef2cd328034796af5e90494195de3fa9298e206400823bb7fbdd8d94cfb8?Expires=1781965229&Policy=eyJTdGF0ZW1lbnQiOlt7IlJlc291cmNlIjoiaHR0cHM6Ly9jYXMtYnJpZGdlLnhldGh1Yi5oZi5jby94ZXQtYnJpZGdlLXVzLzY0ODk5NmZkMTk1YTZmMDlmNjJmYjRiZS81NjlmZWYyY2QzMjgwMzQ3OTZhZjVlOTA0OTQxOTVkZTNmYTkyOThlMjA2NDAwODIzYmI3ZmJkZDhkOTRjZmI4KiIsIkNvbmRpdGlvbiI6eyJEYXRlTGVzc1RoYW4iOnsiQVdTOkVwb2NoVGltZSI6MTc4MTk2NTIyOX19fV19&Signature=MEUCIByXtmAlBQuCoPDDjSAH5s3weAwiF1HXr27fCuJt5TafAiEA32ecz7gtlOip0ihyxGWovScQBxPqEZbYGcdHNdowLWc_&Key-Pair-Id=K1LYXO563TGWFU&response-content-disposition=inline%3B+filename*%3DUTF-8%27%27coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled.zip%3B+filename%3D%22coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled.zip%22%3B&X-Xet-Cas-Uid=public&response-content-type=application%2Fzip&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=cas%2F20260620%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Date=20260620T132029Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=1b86ca8da4ff5c0f317148793a928bb66b8d2c76a5631e6295a009b45fbfe6d2
  [backend] exited with code null
  [download] Completed download of coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled.zip

...

  [backend] Starting: python /Users/user/Applications/Diffusion/app/backend/mac/coreml_server.py --listen-port 8080 --model /Users/user/Applications/Diffusion/app/models/coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled --steps 30 --cfg-scale 7
  [sd-err] /opt/homebrew/Cellar/python@3.14/3.14.2_1/Frameworks/Python.framework/Versions/3.14/Resources/Python.app/Contents/MacOS/Python: can't open file '/Users/user/Applications/Diffusion/app/backend/mac/coreml_server.py': [Errno 2] No such file or directory
  [backend] exited with code 2

I have reviewed the app/backend/mac folder and can confirm that the coreml_server.py does not exist:

How can this be resolved?

[techjarves]

Thanks for the report. You’re right: the CoreML model downloaded fine, but the app was trying to start a missing file:

app/backend/mac/coreml_server.py

I’ve added that missing CoreML backend bridge and updated the Git ignore rules so it will be included in the repo.

I verified the fix with syntax checks:

node --check scripts/serve.cjs
python -m py_compile app/backend/mac/coreml_server.py

To test on your Mac: pull the latest changes, run ./mac.sh, load the same CoreML model again, then try generating a 512x512 image.

[cordschneider]

Nice, thanks!

I've pulled the latest changes and can confirm that the model now loads. However, there is a new error. It appears that the python_coreml_stable_diffusion module is not found:

  [backend] Starting: python /Users/user/Applications/Diffusion/app/backend/mac/coreml_server.py --listen-port 8080 --model /Users/user/Applications/Diffusion/app/models/coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled --steps 4 --cfg-scale 1
  [sd] [coreml-npu] Model: /Users/user/Applications/Diffusion/app/models/coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled
[coreml-npu] Resources: /Users/user/Applications/Diffusion/app/models/coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled
[coreml-npu] Model version: runwayml/stable-diffusion-v1-5
| 1/1 - CoreML resources ready
  [sd] [coreml-npu] listening on http://127.0.0.1:8080
  [backend] READY on port 8080
  [api] Restart backend request: {
  model: 'coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled',
  steps: 20,
  cfgScale: 7,
  sampler: 'euler_a',
  threads: 5,
  use_gpu: true,
  backend_type: 'apple-npu',
  width: 512,
  height: 512,
  vae_tiling: true,
  vae_on_cpu: false,
  flash_attn: true
}
  [backend] exited with code null
  [backend] Starting: python /Users/user/Applications/Diffusion/app/backend/mac/coreml_server.py --listen-port 8080 --model /Users/user/Applications/Diffusion/app/models/coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled --steps 20 --cfg-scale 7
  [sd] [coreml-npu] Model: /Users/user/Applications/Diffusion/app/models/coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled
[coreml-npu] Resources: /Users/user/Applications/Diffusion/app/models/coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled
[coreml-npu] Model version: runwayml/stable-diffusion-v1-5
| 1/1 - CoreML resources ready
  [sd] [coreml-npu] listening on http://127.0.0.1:8080
  [backend] READY on port 8080
  [sd] [coreml-npu-http] "OPTIONS /v1/images/generations HTTP/1.1" 204 -
  [sd] [coreml-npu] generating image
  [sd] /Users/user/Applications/Diffusion/app/backend/mac/coreml_venv/bin/python: Error while finding module specification for 'python_coreml_stable_diffusion.pipeline' (ModuleNotFoundError: No module named 'python_coreml_stable_diffusion')
  [sd] [coreml-npu-http] "POST /v1/images/generations HTTP/1.1" 500 -

[techjarves]

@cordschneider Give me a bit more time. I'll try it on a different Mac. It's working on my M2, so I'll do some more testing.

[sleavitt]

I am encountering this same issue, and the fix for me was to ensure that the version of Python that is used in the venv created in app/backend/mac/app/backend/Mac/coreml_venv was less than Python 3.12, since it appears that you cannot build or install numpy<1.24 on Python 3.12+, since NumPy versions prior to 1.25 rely on distutils in their build configuration, a module that was completely removed from the Python standard library starting in Python 3.12, and Apple's ml-stable-diffusion/setup.py specifically specifies numpy<1.24

I had python 3.14 from Homebrew, and attempting to link it to 3.11 seemed to not work, causing it to revert to the system provided Python 3.9 in Tahoe 26.5.1. This was sufficient for me to successfully generate an image, however.

Perhaps uv or pyenv would be better choices here to create a Python virtual environment with?

[smaquois123]

same problem: No module named 'python_coreml_stable_diffusion'. M5 MacBook Pro, if that helps.

[sometimesnotthatuseful]

Sadly, new install does not work on m4 pro, 24gbs,

Attempted to reinstall and still model will load, but getting generation failed (error 500).

Would appreciate some guidance.

[techjarves]

Thanks for the reports and follow-up testing. I pushed a fix to main for the CoreML setup path.

Please pull the latest code, run the reset script, rerun setup, and test the CoreML model again:

git pull origin main
./scripts/reset/reset.sh
./mac.sh

The setup now uses Python 3.9-3.11 only, recreates a broken/incompatible CoreML venv, pins the compatible NumPy line for Apple's ml-stable-diffusion, and the server now verifies python_coreml_stable_diffusion before showing Apple NPU as ready.

If it still fails, please paste the new console output after the fresh reset/setup run.