AutoMatte

Flame Flame Questions
1

Have been looking at 2026.2 Automatte this morning.

Its not often you get a new feature and immediately think “I could have used this on my last job”! I think the quality is very promising. Looks like the matte can be used (in some cases) beyond just colour correction. Some very good results on a couple of test shots that last job I had to send out for roto, but the Automatte nailed it. It would have been fine for those cases. No waiting about for roto. At the very least a reasonable matte for getting the comp started.

Anyways…… Does ANYONE have any information or ideas about the machine learning ONNX matte models you can load into the Automatte selective?

I assume it would be some pre-trained data sets off Git-hub or something… but the devil is in the details. I’m no programer, I’m less than useless at Linux.

A step by step example from Autodesk would be very helpful. Or if anyone has tried it let me know?

I imagine browsing thru some internet site (git-hub?), which would be the “turbo-squid” of machine learning matte models…. “Oh heck the next job involves Dachshunds…. I need a model trained on dogs” etc etc. Or one that is trained solely on phone screens, or hands, or whatever it may be.

The holy grail of course is “Pants” (@angus ) no more roto’ing the pants.:grinning_face_with_smiling_eyes:

But yep…. Automatte is a great update. Well done Autodesk!

4 Likes
2

Haha! Not so much rotoing pants anymore. More removal of extra limbs some generative AI has just grown for a few frames.

I unfortunately can’t help with your question though. I really need sone ML driving lessons.

3

Hi Justin,

Thanks for the feedback! Yes indeed there is still room for improvement, if you have any use case please let us know.

Regarding the ONNX model, any model that is creating a matte output that is compatible with the Inference node will work. Logik have an installer with models that will work out of the box, so that could be your first place to look for.

Francis.

5 Likes