Reward models?
It would be cool to have reward loras just like before for Wan2.2
I would like to add to the wish list: Control-Camera like before. This model was so good. I can't get the same precision with camera prompting in 2.2 (and thanks to the Wan team for all those nice models! I'm using them since the first EasyAnimate ^^)
Now in training.
I would like to add to the wish list: Control-Camera like before. This model was so good. I can't get the same precision with camera prompting in 2.2 (and thanks to the Wan team for all those nice models! I'm using them since the first EasyAnimate ^^)
https://huggingface.co/alibaba-pai/Wan2.2-Fun-A14B-Control-Camera
I would like to add to the wish list: Control-Camera like before. This model was so good. I can't get the same precision with camera prompting in 2.2 (and thanks to the Wan team for all those nice models! I'm using them since the first EasyAnimate ^^)
https://huggingface.co/alibaba-pai/Wan2.2-Fun-A14B-Control-Camera
Damn! Thank you so much guys!
I would like to add to the wish list: Control-Camera like before. This model was so good. I can't get the same precision with camera prompting in 2.2 (and thanks to the Wan team for all those nice models! I'm using them since the first EasyAnimate ^^)
https://huggingface.co/alibaba-pai/Wan2.2-Fun-A14B-Control-Camera
Hi sorry to ask, is this the same as like Wan2.1-Fun-14B-InP-MPS.safetensors ? I really liked this MPS lora in 2.1, so if there is a 2.2 version coming soon would be amazing :D
Thank you for the Wan2.2-Fun-A14B-Control-Camera ! And good news for the Wan2.1-Fun-14B-InP-MPS.safetensors ! You rock Wan team :)
Sorry for the late release. Reward LoRAs for Wan2.2 are now available at https://huggingface.co/alibaba-pai/Wan2.2-Fun-Reward-LoRAs
Sorry for the late release. Reward LoRAs for Wan2.2 are now available at https://huggingface.co/alibaba-pai/Wan2.2-Fun-Reward-LoRAs
I saw the description. There are reward models specifically for videos like:
https://huggingface.co/KwaiVGI/VideoReward
https://huggingface.co/CodeGoat24/UnifiedReward-2.0-qwen-32b
More on this list as well:
Also, HPSv3 has been released.
We tried to leverage VLM-based reward models into reward LoRA training about a year ago. Training is neither as stable nor as straightforward as with CLIP-based models.
We tried to leverage VLM-based reward models into reward LoRA training about a year ago. Training is neither as stable nor as straightforward as with CLIP-based models.
Oh, interesting.