Shiwei Zhang comments

Results 73 comments of


                                            Shiwei Zhang

Youtube video in README is private

Please refer to [here](https://github.com/damo-vilab/i2vgen-xl/issues/11)

Exception: Failed to init class <class 'tools.modules.clip_embedder.FrozenOpenCLIPTtxtVisualEmbedder'>, with Pretrained weights (models/open_clip_pytorch_model.bin) not found for model ViT-H-14.

Have you placed the downloaded models in the "models" directory?

How to test only the first stage?

In this repo, we only open-source the model for the first stage, but it directly generates 720P videos.

Preview of the image is not available.

Thank you for your detailed feedbacks, we will make the necessary corrections.

How to train I2V

Open-source all the training code of I2V is still in our assessment , but you can refer to the T2V training code and make some limited modifications to train it.

flash-attn install failed

Hi, it seems to be an issue with the version of CUDA, our version is 11.4.

Where is 'workspace/model_bk/model_scope_0267000.pth' in the configs/t2v_train.yaml?

Hi, please refer to [here](https://github.com/damo-vilab/i2vgen-xl/issues/31).

error in inference process (no valid convolution algorithms available in CuDNN)

The previous issue was due to insufficient memory on the GPU card. What GPU card are you using?

error in inference process (no valid convolution algorithms available in CuDNN)

I have only verified on A100. V100 is yet to be verified.

first frame of i2v

Thank you for your interest in our work. Firstly, our I2VGen-XL model has multiple categories. This open-source model can ensure that the first frame of the video is completely consistent...