Yu Lexuan (Blue Hanfu version)- WAN 2.1 T2V 14B

This lora was trained using Wan 2.1 14B T2V model for 50 epochs, 33 frames, 512x512 resolution. It may work using Wan 2.2 in the "low noise" stage, but this was not tested.

All samples are generated using Phr00t's Wan 2.2-All-inOne model due to local compute constraints. It also generates videos a lot faster for folks with xx60 cards.

A good example prompt for the lora looks something like this:

a solo 30-year-old Asian man from ancient times with long flowing hair, a small coronet, and dressed in a blue and white hanfu

For Phr00t's model, use strength 1.0 or close to it. You may need to tone it down using the native models.

Better consistency at 480p resolution, but it will work well at 720p too.

Disclaimer: Users should use models responsibly. My moral stance on the matter means no filtering was applied, and the model can produce if prompted to. generation was disabled on the site, please download the weights and generate externally if you want to.