Hi there. AI rookie question there.
I saw that a Live portrait model using Cattery inference, is available for Nuke
I was wondering if it is usable in flame and if not, how can we “translate” that model to be usable in the inference node of flame 2025 (which I haven’t yet tried)
One more. Is there a way to train models directly in flame. I règlement a video 2 or 3 years ago, in Logiknlive I think where someone showed he had managed to do basic trainings and it was sooooo promising……
Thx
There is no mechanism to translate Cattery models to Flame. We discussed this with ADSK during a Logik Live. Cattery models are encrypted.
However, if the Cattery model is based on an open source ML model (e.g. DepthAnythingV2), then you can take the original ML model and create an ONNX version if it doesn’t already exist. You can check the Logik Portal repository which already includes several. Doing your own conversion does require some python knowledge and a decent size system.
Re: training your own model - no, doesn’t exist yet. Unknown when/if that may be the case. CopyCat makes it very easy to train in user mode. You can train your own models for Flame, but advanced python knowledge is required, and even more GPU time. There’s no UI tool for that.