This is my second actual lora. I even used wrong "model name" while training but since lora is quite alright i don't want to retrain.

It is trained on miniloona's ( @mini.loonaa, @yourminiloona ) Instagram and that blue thingy pictures. Tried to remove any images that face is not visible or where she is not distinguishable. Used auto WD14 auto Captioning. If you check training tags/metadata you will see there are a few positions and pozes that is already in there. I basically use "miniloona" or lora's actual file name (model name is wrong) to trigger it.

Images i am posting are kinda cherry picked. I generated all of them using Automatic1111 with Adetailer (face_yolov8n.pt) and "<lora:miniloona-000024:1>,1girl,portrait,<lora:SDXLHighDetail_v5:0.3>, miniloona" or "<lora:miniloona-000024:1>,1girl,portrait, miniloona" as prompt.

P.S. Feel free to comment, suggest training tips&tricks.

P.P.S. For some checkpoints (although i have manually checked and there is just a few real pictures like that) she has "cream" on her chin so be careful.

edit: i just tested it with comfyui and either i am failing or something is wrong. it creates perfect images in automatic1111 but in comfyui it is as if weights are lowered....