RehgLab / RAVE

RAVE: Randomized Noise Shuffling for Fast and Consistent Video Editing with Diffusion Models [CVPR 2024]
https://rave-video.github.io/
MIT License
269 stars 19 forks source link

local edit #10

Closed rotemshark closed 8 months ago

rotemshark commented 8 months ago

Hi, Thank you for your great work. Which control are you using for the local editing?

Thanks, Rotem

bariscankurtkaya commented 8 months ago

Hi @rotemshark ,

We conducted an ablation study on various types of controls in the paper, as illustrated in Figure 8. The study demonstrates that, despite style differences, these adjustments do not compromise overall consistency (such as changing a train to a cake).

Furthermore, we used depth-zoe as our default control approach in the RAVE paper for all other figures and metric results. For instance, the editing of "a man wearing a glitter jacket is typing" in Figure 1 was performed using this approach.

Thank you for your kind words.

Best, Baris