Zhen Han

I met the same problem with TV-L1 CUDA version. During a while loop, it seems to be run faster at the first several times, but gets slower and slower even...

Sorry, there won’t be further updates since the final versions were already released. For camera movement, you could try layout + prompt to control it with VACE.

I guess you didn't pass mask to the model, so VACE recognized the video are fully changeable. You need to pass a mask video, in which all white are the...

@pftq @lym000000 For anyone who encounters the same problem in ComfyUI: https://github.com/kijai/ComfyUI-WanVideoWrapper/pull/687 The reason of color shifting is probably not setting masks properly.

It seems this issue goes beyond temproal extension task(eg. the above inpainting case). The issue is commoner and the reason must be deeper than I thought. Both VAE and VACE...

Yes, this is the exact way to achieve removal. 1. We recommend to use hull or box shaped mask for removal, to avoid shape leaking. 2. Be sure that the...

Use descriptive prompt instead of instructive prompt.

Hi Niels, Thank you for reaching out and for your thoughtful offer. We are excited to upload our models and create a Hugging Face Space! Running the ACE demo on...

These results look not too bad. If the training set scale is big enough, maybe you should continue training and see. LTX version VACE is experimental. It’s super fast, and...

Layout video with an reference image will be ok. text: descriptive prompt of desired output video; src_video: layout video (recommend boxes if has different shape with reference); src_mask: all-white; src_ref_images:...