Video Generation

Hi, thanks for the great work!

I’m wondering about the current performance of the model on video generation. I couldn’t find any related results or examples in the paper. Also, I noticed in another [issue#16](https://github.com/ByteDance-Seed/Bagel/issues/16) that you mentioned:

> “the model has been trained to generate multi-frames on T2V data, but with a minimal data sampling ratio.”

Would it be possible to include code examples for this functionality in a future update of **inference.ipynb**?

Review inference.ipynb and the referenced issue #16 first to determine what video-generation support is currently exposed. Compare the notebook with the paper's reported results, then document a working multi-frame video-generation example if the existing functionality supports it; done means the example runs and clearly shows the output.

Written by the indexing model from the issue text.