OpenPose input often treated as Lineart/Scribble
Thanks for the amazing work on anima-lllite-any-test-like!
I noticed an issue when using OpenPose. Instead of generating a character in the designated pose, the model often treats the stick figure as Lineart or Scribble, literally rendering the colored lines into the final image.
Would it be possible to release a dedicated Pose ControlNet, or optimize the model to better recognize OpenPose inputs?
Thank you for your hard work!
I believe they did? Anima-LLLite/anima-lllite-pose-1.safetensors
yes,but this one was trained on preview 3 not base model.I also tested it but the result not good.
"I noticed an issue when using OpenPose. Instead of generating a character in the designated pose, the model often treats the stick figure as Lineart or Scribble, literally rendering the colored lines into the final image."
Any test like wasn't trained on open pose data. So it's not that it's bad at pose, it can't do it at all.
And yes, we are eagerly awaiting a controlnet pose retrain on anima base v1. 👍 There is another thread already open requesting it.