Question about generation settings for reproducing RISE-Video

[Qnancy]

Hi RISE-Video team,

Thanks for releasing the benchmark and evaluation code. I am trying to reproduce the reported results for open-source TI2V models such as Wan2.2-TI2V-5B.

I understand from the README that generated videos should be written into the video_path field, followed by frame extraction and evaluation. However, I could not find the exact video generation settings used for the models in Table 1.

Could you clarify the generation configuration used in your experiments?

In particular:

• video length / number of frames
• fps
• resolution
• sampling steps
• CFG / guidance scale
• sampling solver / shift value
• random seed
• whether prompt enhancement or prompt rewriting was used
• whether one or multiple videos were generated per sample
• whether open-source models used their official default inference settings

For example, for Wan2.2-TI2V-5B, should we use the official default generate.py configuration, or did you use a specific frame_num, resolution, guidance scale, or sampling setup?

This would be very helpful for reproducibility and fair comparison.

Thank you!