AI masking throws error

[AndDiSa]

Is there an existing issue for this?

• I checked and did not find my issue in the already reported ones

Describe the bug

cannot create AI object

in the log file you can see message:

2026-03-15 17:59:56.120794549 [W:onnxruntime:, graph.cc:120 MergeShapeInfo] Error merging shape info for output. '/conv_s0/Conv_output_0' source:{1,32,256,256} target:{}. Falling back to lenient merge.
2026-03-15 17:59:56.121053104 [W:onnxruntime:, graph.cc:120 MergeShapeInfo] Error merging shape info for output. '/conv_s1/Conv_output_0' source:{1,64,128,128} target:{}. Falling back to lenient merge.

• additional info:
  I just removed the AI model and re-downloaded it and now at the moment the AI masking is working again ...

Steps to reproduce

1. open mask manager
2. select "add AI object"

Expected behavior

No response

Logfile | Screenshot | Screencast

No response

Commit

No response

Where did you obtain darktable from?

GitHub nightly

darktable version

darktable 5.5.0+623~gdba72f1cda

What OS are you using?

Linux

What is the version of your OS?

Ubuntu 24.04

Describe your system

CPU: AMD Ryzen 9 5900HX with Radeon Graphics (16) @ 4.683GHz
GPU: AMD ATI Radeon Vega Series / Radeon Vega Mobile Series

Are you using OpenCL GPU in darktable?

Yes

If yes, what is the GPU card and driver?

nVidia and Rusticl

Please provide additional context if applicable. You can attach files too, but might need to rename to .txt or .zip

No response

[KarlMagnusLarsson]

AI masking works for me on Linux Debian 13.3 on NVIDIA CUDA.

I also get log file entries when I use it.
2026-03-15 19:04:04.207667507 [W:onnxruntime:, graph.cc:120 MergeShapeInfo] Error merging shape info for output. '/image_encoder/trunk/Concat_3_output_0' source:{4} target:{5}. Falling back to lenient merge.

[AndDiSa]

Yes, currently I am getting the same message as you, but it's a bit different from the one I reported above.

[andriiryzhkov]

These log messages are harmless ONNX Runtime warnings during graph optimization. They don't block model inference, object mask tool should still work fine.

But I will check, maybe I could improve model conversion to avoid this warnings.

[andriiryzhkov]

The MergeShapeInfo warning is an onnxruntime limitation with Hiera's variable window sizes - different blocks produce shape vectors of different lengths for the same op pattern, and onnxruntime's internal shape analysis can't reconcile them. It's cosmetic and doesn't affect results.

[andriiryzhkov]

@AndDiSa are you still having problem creating AI object mask?

[AndDiSa]

@andriiryzhkov no, not any longer. It was very strange as it was working for a time and then suddenly I got these MergeShape-Messages and in addition on the screen there was a message like "object mask could not be created"
I then deleted the model in the AI-Manager and reloaded it completely new.
After that the issue was gone.

[AndDiSa]

It's currently not reproducable, to I close the issue