I'm having some issues with your workflow for image to video.
When I try to generate long videos, such as 30-second videos, the last few frames of the video are usually problematic, with severe artifacts. Like this:
What happen? How can I solve this problem?
When I try to generate long videos, such as 30-second videos, the last few frames of the video are usually problematic, with severe artifacts. Like this:
Regular Image to video workflow? LTX-2 is made for 10-15 second videos or so.
If you go far beyond that, things will get strange, model is not trained for that
(you can try my other workflow for long video, that works differently, by rendering 10-15 seconds videos and then another and another, and sticking them together seamlessly)
Or if you did use the long video workflow, try different seed in the group that caused issues.
There will be glitches from time to time, i seen that too, mostly big titles on screen...
Its a bit of an experimental workflow, will see if anything can be improved upon ;-)
Thanks, I'll try it.
What happen? How can I solve this problem?
It could also be the upscaler.
LTX released a new version of the spatial upscaler, version 1.1 to fix some of this issue at longer videos.
I havent tested yet, but do give it a try ;-)
Yeah! I tried the new spatial upscaler model, and this problem has been solved.
ah good good ;-)