Training code

[shawnFuu]

Thanks for your great work. I wonder whether you open-source the training code. Can I transfer your method to other t2v models?
Looking forward to your early reply.

[JianhongBai]

@shawnFuu Thanks for your interest. The model used in our paper is an internal T2V model that cannot be released due to company policy restrictions. We are in the process of migrating our method to CogVideoX and retraining it. We will open-source the training code and checkpoint in a month. In addition, you can refer to https://github.com/KwaiVGI/SynCamMaster?tab=readme-ov-file#3-code-snapshot to transfer the proposed method to other T2V models.

[HyoKong]

@shawnFuu Thanks for your interest. The model used in our paper is an internal T2V model that cannot be released due to company policy restrictions. We are in the process of migrating our method to CogVideoX and retraining it. We will open-source the training code and checkpoint in a month. In addition, you can refer to https://github.com/KwaiVGI/SynCamMaster?tab=readme-ov-file#3-code-snapshot to transfer the proposed method to other T2V models.

Hi, may I ask the VRAM cost for training the model? Thanks!

[wz0919]

@shawnFuu Thanks for your interest. The model used in our paper is an internal T2V model that cannot be released due to company policy restrictions. We are in the process of migrating our method to CogVideoX and retraining it. We will open-source the training code and checkpoint in a month. In addition, you can refer to https://github.com/KwaiVGI/SynCamMaster?tab=readme-ov-file#3-code-snapshot to transfer the proposed method to other T2V models.

Thank you so much for your amazing work! I was wondering if there are any updates at the moment about the checkpoints and the training codes (Looks like one month passed)? I’m really excited to get started with the model. Thanks again!

[JianhongBai]

@shawnFuu Thanks for your interest. The model used in our paper is an internal T2V model that cannot be released due to company policy restrictions. We are in the process of migrating our method to CogVideoX and retraining it. We will open-source the training code and checkpoint in a month. In addition, you can refer to https://github.com/KwaiVGI/SynCamMaster?tab=readme-ov-file#3-code-snapshot to transfer the proposed method to other T2V models.

Hi, may I ask the VRAM cost for training the model? Thanks!

Thanks for your interest. When training a 2-view generation model on the CogVideo-5B using 8 GPUs with a batch size of 1, the VRAM usage per GPU is approximately 70GB.

[JianhongBai]

@shawnFuu Thanks for your interest. The model used in our paper is an internal T2V model that cannot be released due to company policy restrictions. We are in the process of migrating our method to CogVideoX and retraining it. We will open-source the training code and checkpoint in a month. In addition, you can refer to https://github.com/KwaiVGI/SynCamMaster?tab=readme-ov-file#3-code-snapshot to transfer the proposed method to other T2V models.

Thank you so much for your amazing work! I was wondering if there are any updates at the moment about the checkpoints and the training codes (Looks like one month passed)? I’m really excited to get started with the model. Thanks again!

Thank you for your attention. We have found that the training on CogVideo-5B is extremely slow and requires much more debugging work than expected. We are working hard on the migration, which may take additional time.