add <a rel="ugc" href="https://civarchive.com/models/723360/ai-styles-dump-animaillustriousrouweinoob?modelVersionId=2856394">AI styles dump</a>recent dataset comes from <a rel="ugc" href="https://civarchive.com/models/2495369/kirazuri-anima?modelVersionId=2898376">Kirazuri (Anima)</a>and <a rel="ugc" href="https://civarchive.com/models/1252497/photo-background-2d-compositingyuan">Photo Background - 2d Compositing</a>and <a rel="ugc" href="https://civarchive.com/models/2385278/animayume?modelVersionId=2851312">AnimaYume</a><hr />AnimaYume and Kirazuri (Anima) both updated the llm_adapter during training. If you directly merge them, when using certain Danbooru tags, the images will appear to have an excessively high CFG level. I processed this part using a more refined method, making prompts such as "photo background" and "Aesthetic tags" from the Kirazuri (Anima) model effective without having to lower the CFG.The trigger words from the AI styles dump also responded positively.It is recommended to use the sa_solver_pece sampler and terminate the SDE earlier to obtain more detailed images.

Files with SHA256: 0e9171f18b542e8b6d6b6a3d1a144a23f50c5a15a2c64f56247cd359c7c9bffa