Compare with InfiniteTalk

[00Neil]

Good job! I wonder how the quality of the results compares to that of InfiniteTalk ? How long can a video be generated at most while ensuring quality?

[caijiaran1993]

Our paper was submitted before InfiniteTalk was released, so a direct comparison was not included. Subsequently, we conducted internal comparisons and observed that our method has certain advantages in facial expression diversity and motion richness, while achieving comparable lip-sync accuracy. In terms of performance, without using 4/8-step denoising or quantization, the inference performance should be similar.

[ghx2757]

您好，是否支持使用4/8-step的lora蒸馏模型？如lightx2v或者fusionX，如果支持的话您使用的那个模型呢？ # @caijiaran1993