Is there any plan to open source the imae-to-video model and related code?

[1343744768]

Hi, Thanks for your amazing work! May i ask is there any plan to open source the image-to-video model and related code?

[SHYuanBest]

Thanks for your interest. Our current code supports both image and text inputs simultaneously. When only an image is input, it functions as an image-to-video model, but the quality of the generated video may degrade.

[wangqiang9]

Thanks for your interest. Our current code supports both image and text inputs simultaneously. When only an image is input, it functions as an image-to-video model, but the quality of the generated video may degrade.

I have a question. When only an image is input, how should the input text be set? Can it be set to null? How does text cross attention work?
Thanks a lot.

[SHYuanBest]

the input text will be set to "" (Empty string). At this time, all T5 tokens are padding tokens, which means they do not function as text. Of course this will drastically reduce the quality of the generated video

[wangqiang9]

the input text will be set to "" (Empty string). At this time, all T5 tokens are padding tokens, which means they do not function as text. Of course this will drastically reduce the quality of the generated video

Thanks.