视频生成结果乱码

[interestingzhuo]

python tools/t2v_infer.py 结果错误，

[HelloJianHan]

@interestingzhuo We’ve updated the model architecture and revised the text embeddings from unidirectional T→V attention to bidirectional attention. As a result, the old models are no longer usable. Fortunately, I’ve completed SFT for the updated T2I and T2V models. Feel free to pull the latest code and give it a try. Good luck!