Video generation takes too long even with GPU

[Tabz05]

Greetings,

I am Mohammad Tabish Shamim, an MSc Artificial Intelligence student at the University of Southampton.

For my MSc dissertation, I am researching on zero-shot text-to-video and I have based my research primarily on your research paper titled "Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generators". I was trying to run the code of Text2Video-Zero model which is available on GitHub. However, for an input text prompt, the video generation process goes on indefinitely and an output video does not get generated.

I would like to bring to your attention that I was running the code using GPU, i.e., torch with cuda. I selected the CompVis/stable-diffusion-v1-4 model. Moreover, I did not make any changes to the code. In order to get an output video quickly, I had reduced the video length to 1 second and set the merging ratio value to 0.9. However, none of the attempts proved to be fruitful; the video generation process proved to be indefinite. Am I missing something? For your reference, I have attached pictures of the configurations.

I would be grateful to you if you could look into this concern of mine.

Looking forward to hearing from you at the earliest.

[Tabz05]

Greetings,

Kindly resolve my query. Looking forward to hearing from you at the earliest.

Regards

[Tabz05]

Greetings,

Looking forward to hearing from you at the earliest.

[contact2saxena]

Same with me