Update: using the pony version with an anime style lora fixes most of the problems and also improves the style. I updated the image gallery with new images with better example prompts. (a strenght of 0.9 works best to preserve the style)

Update: added a pony version. I'm not very satisfied with that version, but it does work. I don't know what went wrong but it tends to generate images without a face. I'll update if I can make a better version. Recommended strength: 0.8.

A style lora trained on male characters from the Winx series to capture their animation style. Not trained on a specific character and not trained on female characters.

I have made 2 versions, trained with the same dataset.

One trained with AnytingV3, and one trained with the 1.5 base model

The version trained on AnythingV3 works better with anime models.

recommended strength: 0.75