First-frame + Last-frame workflow for video generation?

#12
by Dodome - opened

In older video workflows, we could provide a starting image and an ending image, and the model would generate the frames in between.
Is something like that possible with LTX-2 right now? Is there any workaround or planned support?

@Dodome yes, it's absolutely possible

see this workflow (native Comfy)

https://gist.github.com/kabachuha/dafd6952bdc00050b4d6b594d11bec6c

@kabachuha Thank you, it worked.

I just tried it. It made the most jank output ever. Instantly abandoned the first frame input and the last frame is something out of lala land. Im not sure what is going on here but that wonder.json workflow does NOT work.

I just tried it. It made the most jank output ever. Instantly abandoned the first frame input and the last frame is something out of lala land. Im not sure what is going on here but that wonder.json workflow does NOT work.

Yeah, it's not yet supported properly but it works play with the strength of frame_idx.
Though it won't give you good results always, for me it worked on some tests.

Sign up or log in to comment