Redistribution is prohibited.
ABOUT THE MODEL
Jinhsi from Wuthering Waves
20260326v0.1
Trained on Anima Preivew 2
At the moment, I’ve found that training LoRa on Anima Preview2 still comes with quite a few issues. Custom tags are not easily learned, and the concepts I want to control tend to get incorrectly mapped to other inherent tags—or even to tags that don’t appear in the captions at all. This is likely because the text encoder isn’t being trained. Even when I tried training with an LLM adapter simultaneously, there wasn’t any noticeable improvement.
This problem becomes even more pronounced in multi-concept training, where different concepts tend to interfere with each other. It also makes it difficult to clearly separate them using custom tags. I’ve encountered similar issues before when training LoRa on Chroma as well.
For now, I don’t have a good solution to this problem. I’m not sure whether future versions of Anima will improve the training behavior. Although Anima Preview2 seems capable of learning more information from the dataset, it still feels relatively slow in terms of fitting speed compared to models in the SDXL series.
But it has to be said that Anima’s understanding of concepts and semantics is consistently impressive. The model itself has already learned a wide range of concepts, and motion that are difficult to achieve with SDXL series can be handled by Anima with ease.
Looking forward to better performance in the next version of Anima.
Features
3+ outfits
Anime style
Issues
There’s interference between clothing concepts — the default outfit is overfitted, causing weaker performance on the other outfits.
Control over clothing details is also not stable enough.
When the face becomes blurry, consider using ADetailer to fix it
TRIGGER WORDS
Default
j1nhs1, silver hair, long hair, blunt bangs, twintails, light blue hairbow, gold hairclip, silver eyes, black underdress, white overcoat, cyan cape, armband, wristband, white thighboots, Under Dress
j1nhs1, silver hair, long hair, blunt bangs, twintails, light blue hairbow, gold hairclip, silver eyes, black underdress, barefoot,Hanfu Dress
j1nhs1, silver hair, long hair, blunt bangs, twintails, light blue hairbow, gold hairclip, silver eyes, blue long chest-skirt, sheer-robe, orange waist ribbon, orange sash around arms, Requirements
Checkpoints: Anima Preview 2
Recommended Tools: ComfyUI, Forge Neo
Recommended Step: 20~30
Recommended weight: 0.6~1
Recommended size: 768*1344, 832*1216, 896*1156, 1024*1024, 1344*768, 1216*8
32, 1156*896
I’ve set the gallery moderation of this post to “R”, so explicit content may not appear on this page.