Settings
Look, with the previous settings, it turns out like this - temp 16 - 3 seconds. By setting the resolution lower on the 768 model, the picture begins to disintegrate into artifacts from the middle of the process. I don't know why, but temp 16 is still producing 3 seconds of video and the time has already taken 14
minutes to generate in the ComfyUi (on a 3060/12 graphics card)
Hi! What is the setting of these two videos? By the way, you should not use the 768p version checkpoint to generate low-resolution (e.g., 640x384) video.
You should try the 384p version checkpoint to generate the low-resolution (e.g., 640x384) video.
a similar effect happens when merging latents and continuing to sample from an incorrect start step (flux dev latent blending "non video"), the denoise essentially falls short, and in the opposite (an extra step) creates a smoothness to an output. i can see the similarity :)
appologies, as random as this may be, i hope this information helps relate a phenomenon :)
Hi! What is the setting of these two videos? By the way, you should not use the 768p version checkpoint to generate low-resolution (e.g., 640x384) video.
Yes, thank you! I realized this thanks to experiments and trials) I just wanted to check) The only pity is that when generating videos on temp 16 in 1280x768 resolution, the generation time is about 23-25 minutes (16/16 [23:40<00:00, 88.76s/it])