SteadyDancer workflow and fp8 or gguf?
This model is better than Wan Animate 2.2
Implementation is work in progress, but it is really not better overall than WanAnimate though:
- it has no face control at all
- it has no long generation method (can somewhat work with context windows, but it's not great)
- it has no built in inpaint feature
Wait, seriously? So it’s a disappointment? Still, thanks, Master Kijai, for the early heads-up. What I like is that it maintains the first frame, though.
Wait, seriously? So it’s a disappointment? Still, thanks, Master Kijai, for the early heads-up. What I like is that it maintains the first frame, though.
Yeah it's not bad at all, just limited usability without the above features.
I pulled down the SteadyDancer git project, and it's a bit of a mess.
Would you mind sharing how to spin up the Wan21_I2V_SteadyDancer_fp16.safetensors model in ComfyUI?
I pulled down the SteadyDancer git project, and it's a bit of a mess.
Would you mind sharing how to spin up the Wan21_I2V_SteadyDancer_fp16.safetensors model in ComfyUI?
It's not fully finished as it's missing their pose detector and the negative pose augmentor, but you can use it with lightx2v and cfg 1.0 now by just updating the WanVideoWrapper, there is simple example workflow.