Recommended Use: 0.65 weight at 1-3cfg for likeness. Use higher cfg for complex prompts but inpaint/adetailer face at 1-3 cfg.This Lora can do just about anything. It helped that she cosplays in many different hair colors, styles and costumes. And I found that using batch size 1 in training helped the likeness the most.While I was trying to train a Lora that looked good at 1.0 weight, I notice the likeness would never be there. So I overtrained a ton, the likeness at 1.0 would then be interesting because it would try really hard to recreate images to the point of being very fuzzy, but then lowering the weight would keep the likeness minus the fuzziness. Anyways I discovered you can just look up my settings in webui itself so I won't write it here but knowing the logic helps.The bonus with using such low weight is that it becomes very flexible while retaining amazing likeness. Big success.

Rakukoo (Very Flexible Lora)

Rakukoo-40r2e.safetensors

40 repeat 1epoc

Files with SHA256: 0c9e70063b86e2bc8909e29c32b608d994efba7d2c0c3f7200c89c393b87e1a7