This is an experimental LLLite Regional Controlnet model for Regional conditioning for Anima. <a target="_blank" rel="ugc" href="https://huggingface.co/Sen-sou/Anima-LLLite-Regional-Controlnet">https://huggingface.co/Sen-sou/Anima-LLLite-Regional-Controlnet</a> <h3 id="usage">Usage</h3>Use a color mask image as the conditioning input. Any color can be used to define a region. There is no fixed palette or strict RGB requirement. The mask background should be white. Using simple solid colors (red, green, blue, yellow, etc.) for different regions is recommended for clarity. The model was trained using manually masked conditioning images and therefore expects clearly separated regions. Install <a target="_blank" rel="ugc" href="https://github.com/kohya-ss/ComfyUI-Anima-LLLite">https://github.com/kohya-ss/ComfyUI-Anima-LLLite</a> in comfyui to use the model.Controlnet end step value is recommended to keep at 0.45.A higher controlnet strength can be used if the subject is leaking through the mask bounds <h3 id="prompting">Prompting</h3>Normal prompting works fine.However, the model currently cannot determine which prompt corresponds to which colored region automatically. It only receives the region mask as additional conditioning and does not perform explicit prompt-to-region matching.To guide concepts into specific regions, use spatial prompts such as:<ul><li>"girl on the left, cat on the right"</li><li>"character in the foreground, city skyline in the background"</li></ul>Combining this model with attention masking methods such as Forge Couple or the 'Attention couple' node by <a target="_blank" rel="ugc" href="https://github.com/pamparamm/ComfyUI-ppm">https://github.com/pamparamm/ComfyUI-ppm</a> to get better results <h3 id="training">Training</h3>Trained on 580 images, 2 repeats, batch size 4, for 4400 steps Conditioning images were manually masked for each image Captions were generated with the help of <a target="_blank" rel="ugc" href="https://huggingface.co/SmilingWolf/wd-eva02-large-tagger-v3">wd-eva02-large-tagger v3</a> and <a target="_blank" rel="ugc" href="https://github.com/pythongosssss/ComfyUI-WD14-Tagger">https://github.com/pythongosssss/ComfyUI-WD14-Tagger</a> custom node Trained using <a target="_blank" rel="ugc" href="https://github.com/kohya-ss/sd-scripts">https://github.com/kohya-ss/sd-scripts</a><h3 id="limitations">Limitations</h3>Because most training images consisted of close-up character compositions, generations involving distant subjects may not strictly adhere to the provided mask boundaries. Support If you want to support future training, you can support me via Ko-fi <a target="_blank" rel="ugc" href="https://ko-fi.com/sensou">https://ko-fi.com/sensou</a>

Files with SHA256: f8cdc9897a4eaa50b0f9bc18dda098fb096eeb64df001589492b9774d0cf5cc5