isclaimer

Although base HunyuanVideo knows generic anime style well without need for LoRAs and has some knowledge about Studio Ghibli's art style, the latter is not consistent, very prompt-dependant and can sometimes fall back to realistic style. And the shading, palette and linework can be quite different. So by making this LoRA I wanted to try to reinforce the Ghibli art style for HunyuanVideo.

This is the third version of the LoRA. The first two versions were not successful, I did not publish them. And this is not final version, I am working on improving it.

upd. 05/01/2025 Done training v.0.4 with musubi-tuner, but it was worse than v0.3, so I won't publish it (and will use diffusion-pipe for v0.5).

upd. 08/01/2025 Training v0.5 is progressing, but slower than I’d hoped. Perhaps I was too optimistic about the speed of training 750 clips at a resolution of 512x* (max frames 33) on an RTX 3090. And at epoch 11, the results are still not good enough.

upd. 09/01/2025 Taking a break to train another anime LoRA (and to check some assumptions meanwhile), but will resume v0.5 training right after.

I am still figuring out how to train HV and still do not know the best way to prompt it, etc., take that into consideration.