Trained with 78 imagesIf you like my work, please consider supporting me: <a target="_blank" rel="ugc" href="https://ko-fi.com/suzushi2024">https://ko-fi.com/suzushi2024</a>I am open to ideas and suggestions on characters lora, models etc.Also built a free site to generate image and host models, more models are coming soon,Try it out on : <a target="_blank" rel="ugc" href="https://miyukiai.com/">https://miyukiai.com/</a>The Zimg lora is experimental. Prompt can be a bit hard I am using xml to see how the model adapts to it.The prompt style goes as the following.&lt;character&gt;&lt;/character&gt;&lt;quality&gt;&lt;/quality&gt;&lt;danbooru_tags&gt;&lt;/danbooru_tags&gt;&lt;caption&gt;&lt;/caption&gt;Since the base model didn't really learn a lot of danbooru tags, the caption in natural language remains more important. But Z image generates anime style image pretty good already.

Files with SHA256: 499ead3aa8cf29637197697534e84ca01148f9113ffd03aa857ce74efb598b74