Trigger word: Setsuna
Under the Wan T2V 14B model, the output is closer to the original anime aesthetic, maintaining a more faithful animated look. However, FusionX is generally recommended for its enhanced cinematic feel, especially in scenes. While characters may appear slightly sharpened, the overall atmosphere and visual depth are significantly improved.
All videos can be generated using only the trigger word "Setsuna" — no character description is necessary (though clothing details can be added for variation).
When using FusionX, make sure to include the tag "Anime style" to avoid potential real-life (photorealistic) generations.
FusionX performs well with both characters and environments, and has shown solid results in tests.
When training a LoRA for an anime character, it often feels more like creating a style LoRA, especially since WAN’s base model is more oriented toward realism. Since you usually can’t generate an entire video using just a single character, I suggest training on the full anime series to ensure enough variety in the data. When I trained this LoRA, I included a lot of twilight scenes featuring Setsuna, which helped it generate some nice sunset-themed anime visuals even when used in fusion.
Buy me a coffe: https://ko-fi.com/dreamweebs
Planning to train for White Album 2 full series, but not sure how that will go as I want to keep all the main characters.