The released I2VGen-XL model seems like a single stage model

[junbangliang]

Hi,

Thanks for sharing the code for I2VGen-XL model.

From closer inspection of the unet_i2vgen.py file, it seems to me the released I2VGen-XL model is a single stage model where it takes in both image and text at the same time, without performing two stage processing as claimed in the paper. Is this the case?

Thanks!

[Steven-SWZhang]

Yes, in this open-source project, we have only released the model for single-stage high-definition video generation that can fully preserve the content of the input image, for the convenience of academic use. As for the multi-stage video generation models, we currently do not have plans to open-source it. Thank you for your interest in our work.

[junbangliang]

Hi,

Thanks for your previous answer, I have another question on the released single stage I2V model. Is it also trained with 35 million videos and 6 billion images?

Thanks!

[Steven-SWZhang]

Hi,

Thanks for your previous answer, I am another question on the released single stage I2V model. Is it also trained with 35 million videos and 6 billion images?

Thanks!

Yes.