TianxingWu / FreeInit

[ECCV 2024] FreeInit: Bridging Initialization Gap in Video Diffusion Models
https://tianxingwu.github.io/pages/FreeInit/
MIT License
481 stars 22 forks source link

work with svd? #1

Open Laidawang opened 9 months ago

Laidawang commented 9 months ago

can it be used with svd(stable video diffusion)?

TianxingWu commented 9 months ago

Thanks for your interest. Yes we have conducted some preliminary explorations on SVD, and found that FreeInit can indeed enhance the consistency in several cases. However, since SVD noise schedule is largely different from SD, some adjustments might be needed for optimal effectiveness.

LiuShiyu95 commented 9 months ago

@TianxingWu Thank you for your great work. Have you tried inversion the video in SVD to get the noise and then reconstructing the original video?

PHOENIXFURY007 commented 8 months ago

Any updates on this ? Can it be used with I2Vgen-XL ?

howardgriffin commented 4 months ago

any update?

TianxingWu commented 4 months ago

@howardgriffin A recent paper called ConsistI2V seems to be working on similar ideas, but with specific designs for I2V tasks. Maybe you can check their work first :)