How these videos are made? they seem damn realistic!!

my guess is theyre probably starting from a really good still/reference image instead of generating the whole thing from text.

once the first frame already looks real, image-to-video only has to handle the movement. then you generate a bunch of short clips, keep the cleanest few seconds and edit them together.

thats probably why these look way better than the usual one prompt = one full video stuff.
 
Back
Top