Correction to Cited Paper Title and Suggestion for New Baseline Method (Concat-ID-Wan-AdaLN)

[yongzhong-ruc]

Thanks for your excellent work.

I noticed that the title of the cited paper has a small issue — the correct version should be "Concat-ID: Towards Universal Identity-Preserving Video Synthesis".

The BibTeX format is:

@article{zhong2025concat,
  title={Concat-ID: Towards Universal Identity-Preserving Video Synthesis},
  author={Zhong, Yong and Yang, Zhuoyi and Teng, Jiayan and Gu, Xiaotao and Li, Chongxuan},
  journal={arXiv preprint arXiv:2503.14151},
  year={2025}
}

In addition, I updated Concat-ID with the Wan2.1-T2V-1.3B video model recently. Would it be possible to add a baseline method named Concat-ID-Wan-AdaLN (https://github.com/ML-GSAI/Concat-ID?tab=readme-ov-file#concat-id-wan-adaln)?

Thank you again for taking our work as a baseline.

[SHYuanBest]

Thanks for your great interest! I used the correct bib, but the latex template automatically converted it to lowercase. And I will add the great work Concat-ID-Wan-AdaLN to OpenS2V-Eval later.

[yongzhong-ruc]

Thanks for your great interest! I used the correct bib, but the latex template automatically converted it to lowercase. And I will add the great work Concat-ID-Wan-AdaLN to OpenS2V-Eval later.

Thank you very much.

[SHYuanBest]

Alternatively, you can use OpenS2V-Eval-Human to sample videos and send them to us for quick evaluation updating, we can run the eval scripts for you. (I'm currently occupied with other tasks)

[yongzhong-ruc]

Alternatively, you can use OpenS2V-Eval-Human to sample videos and send them to us for quick evaluation updating, we can run the eval scripts for you. (I'm currently occupied with other tasks)

Okay, I'll give it a try.

[SHYuanBest]

@yongzhong-ruc hi, we just upadte the result, click here and here for details. It seems that there is a great improvement in subject consistency (concatid-wan-adaln), but due to the limitation of the model being only 1.3B, the basic quality of the video is slightly worse than the old 5B version (concatid-cogvideox-5B).

[yongzhong-ruc]

@yongzhong-ruc hi, we just upadte the result, click here and here for details. It seems that there is a great improvement in subject consistency (concatid-wan-adaln), but due to the limitation of the model being only 1.3B, the basic quality of the video is slightly worse than the old 5B version (concatid-cogvideox-5B).

Thank you very much!

[SHYuanBest]

@yongzhong-ruc Hi, we've updated the OpenS2V-Eval-v1.1 Leaderboard. Concat-ID-Wan-AdaLN has achieved some impressive results with just 1.3B parameters.