Can I use the framework to distill a bidirectional HunyuanVideo(-I2V) model?

#4
by SnowflakeWang - opened

Hello, thank you for your great work!
I am inspired by your usage of self-forcing to accelerate bidirectional Wan model! Another open-source model, HunyuanVideo(-I2V) is also a bidirectional one. I wonder whether I can use your training framework to accelerate my HunyuanVideo(-I2V) model? Can I directly train a LoRA on a specific small dataset for HunyuanVideo(-I2V) to accelerate the generation?

Thanks a lot!

Sign up or log in to comment