Text prompt and motion bucket questions

[domejiraphon]

Hello, and thank you for sharing your work—very cool!

I've been trying the demo and, from what I've learn from your paper, your model can control the generated video using a single image, a text prompt, and a specified trajectory. However, I noticed that there's no option to input a text prompt in your demo. Where do you condition on the prompt?

I also came across the term "motion bucket" in the demo. Could you clarify what that refers to?

[chenfei-wu]

Hi @domejiraphon,

DragNUWA 1.0 (in original paper) supports text, image, and trajectory as input controls.
DragNUWA 1.5 (in this implementation) supports image and trajectory as input controls since it uses SVD as the backbone.

Motion buckets can be used to control the motion of the generated video.

Thanks.