Feedback on the V8 model.
(Translated by Google) The issue with incomplete denoising for 1-2 frames across all models has been resolved. The V8 I2V NSFW model's facial consistency is poor, likely due to the effects of the fused LoRa. The normal V8 model performs perfectly, and the excessive noise issue with the V7 has been resolved. However, the LoRa intensity for WAN 2.1 needs to be increased to 1.35-1.4 to achieve the desired effect (lower intensity will not be noticeable), and higher intensity will also affect facial consistency. Also, I'm not sure if it's my workflow, but the videos generated by V8.1 T2V are very slow (mostly, the generated video is a fixed shot with almost no subject motion). I've seen suggestions that setting the intensity of the high-noise acceleration LoRa（Lighting） to 2 can resolve this issue. Thank you for your work!