This is mostly experimental, if you want better experience and are not pushing the limits you will be likely more happy with <a target="_blank" rel="ugc" href="https://civarchive.com/models/2727446/moefussion-30-anima-class">https://civarchive.com/models/2727446/moefussion-30-anima-class</a> Moe Fussion Krea Moe Fussion focuses on providing wide spectrum of models suitable for different purposes. You are free to use these models however you want to what Krea License allows. Please read a rest to know how to use these models. General Info: This is a whole new for now experimental generation of models based on Krea, better prompt adherance, composition, knowledge, details than SDXL based models like Illustrious. The ecosystem is new so you may not find Controlnets and some LoRAs yet. This model was made to bring anime and booru prompting to krea along with default anime style. Before you use this model make sure you have fully updated ComfyUI or SwarmUI, I do not guarantee it will works somewhere else. Cfg: around 3.5 Steps: around 50 (or use turbo lora) Sampler: Euler (do not use euler a or mostly any other sampler, I observed degradation when plain euler is not used, higher steps compensate tho) Scheduler: Simple Negative prompt: bad quality, monochrome, simple background, (or none tbh, not really needed) Prompting style: use danbooru tags with combination of natural language, longer prompts may perform better, it is as well possible to use just tags or just natural language it will work just fine. For artists not really supported but you may try These are not absolute rules how to use the model, feel free to experiment, for some specific scenarios different configuration may be better.

Files with SHA256: 025d6adcb0a7e1d7e542acece5f6aa0f3b6134cbfd66e043387b9a7e754c7c0f