Anime Style for Hunyuan

Hunyuan struggles with anime aesthetic and motion. This lora helps tune your generations for a modern Japanese animation style.

No special trigger or prompting required, but adding "anime style" may help significantly.

If you have questions, or want to commission a lora, join the discord.

Recommended Settings:

Resolution: This lora was trained on 420px images and videos of assorted aspect ratios. It should work on generations of any resolution. I've had great success with 432w x 240h, then 2x upscale. See below for more info.

Frames: Please experiment. 24 and 48 frames have worked for me, longer generations are trickier to make work, but that's mostly because it takes longer to preview. Feel free to try out a 1-frame preview to see if you need to adjust the strength before continuing.

Steps: 10 worked for me, but please keep reading about upscaling.

Upscaling: Hunyuan gives better results when running a second upscale pass, it generally cleans up motion and adds better detail. It can help fix things that didn't follow the prompt in your initial gen, such as moving/talking mouths even when you have "closed mouth" in the prompt. I like to do a 1.5x or 2x upscale once I'm satisfied with how my initial video looks.

Transparency:

This lora was trained on 47 images and 27 video clips. It took $5 worth of Runpod credit on an RTX 4090. I will never ask for your buzz or donations, but I will ask you to leave feedback in the comments so I can continue to improve my work.