(Solution) FLF2V last frame doesn't match input image
#110
by FartingBackwards - opened
I struggled with this for longer than I should have, so hopefully this can help until the workflows are updated.
Last frame is not anchored in the second sampler pass, so the model is free to invent whatever it wants at the end of the generation. The fix is to include an LTXImgToVideoInPlace for the last frame, as in the first pass.
Change this:
To this:
And the generated video's last frame will finally match the input keyframe.
FartingBackwards changed discussion title from (Solution) FLF2V last frame doesn't match input frame to (Solution) FLF2V last frame doesn't match input image
Yes that makes perfect sense, and if the workflow don't have that already, thats a bit of an oversight for sure. Too many workflows and not enough coffee ;-)
Will update