The first frame of result video seems very blur
Hi, thank you for your great work!
When testing this checkpoint, I noticed that the first frame of the generated video is noticeably blurred, and the text-to-video semantic consistency is not ideal. I’ve tried different samplers, but the results are quite similar in terms of both blur and semantic mismatch.
I'm using the following ComfyUI workflow for testing:
🔗 https://www.runninghub.cn/workflow/1953377715611205634
Has anyone encountered similar issues? Or are there any suggestions on how to improve the first frame clarity and semantic alignment?
Thanks in advance!
Forgot to add all these models are I2V, and not T2V.
This ComfyUI workflow which I use is indeed I2V: https://www.runninghub.cn/workflow/1953377715611205634, And the results are very strange. Can you show me the correct workflow to make this checkpoint work?
FYI 1-2 frames of noise is expected with merges I2V V2+ "all in one" merges. I recently put out V8 which significantly improves starting noise, but doesn't get rid of it completely. This FP16 version was V5, I believe.