This is a finetuned (and merge) model of sdxl, based on a mixture of checkpoints such as CounterFeitXL, AIO anime, SDXL 0.9, finetuned and further merged. It's a work in progress, implying that there might be further changes or updates.

The model is trained on 300k anime images with danbooru tags for language understanding, and then applied checkpoint merging for a more aesthetically pleasing result.

There are no required tags or embeddings for the model to work, but a mixture of danbooru tags, natural language, and quality tags could be beneficial. Also the model might require a higher CFG value to work have better results.