txt2vid: dynamically change "Sampling Steps" based on a preview video render. #1076
cstueckrath
started this conversation in
Feature suggestions
Replies: 1 comment 4 replies
-
Can i ask layman's question? How can i get this feature in the current web ui? I dont see any txt2vid panels / sliders/ settings in the UI. Maybe there is another .cmd file i need to run to get to it? I know there is a repo out there that focuses on the video generation but its a separate repo from this and its a bit clunky and not many options to tweak in the UI |
Beta Was this translation helpful? Give feedback.
4 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
currently txt2vid has two relevant sliders named Sampling Steps and Inference Steps.
What Inference Steps does is comparable to Sampling Steps in txt2img: Quality and detail of the image
What Sampling Steps does is to control the "frame rate" of the video. If you set this to 64 and render a video and then set it to 128 and render from the same seed, the resulting video will have 1 frame between each frame of the first video.
If you double the rendered frames, the beginning and end of the videos will be the same (+/- a frame)
We can use this to skip through a video and see, what it will look like before letting it run over night with high quality settings.
But if we could change the settings while running, we could even cherrypick the part of the video we want.
a) calculate frames at very low sampling steps and inference steps (1) to reach beginning frame
b) iterate through until beginning frame is reached
c) change sampling steps and inference steps mid flight to user configured values (e.g. 128/64)
d) render until end frame is reached
Beta Was this translation helpful? Give feedback.
All reactions