G-U-N/Gen-L-Video

Adapting existing ModelScope architectures to generate long videos

Opened this issue · 1 comments

Hi, I've been trying this repo out the last few days and it's really amazing. I wanted to see what is the maximum length to which I can stretch video generation without fine-tuning. I tried a lot but it doesn't seem like there's any script to adapt an existing short t2v model for long generation yet. I was wondering if you can guide me on how to achieve this, just even a pseudocode approach would be great. If it ends up working, I can submit a PR in case you're interested in integrating it with the rest of the code.

Did you ever get anywhere with this?