LoRA trained on 852 mixed resolution images of Akai Riot's work. 6 epoch and 12 epoch versions are provided as they seem to both work at times (sometimes 12 epoch is way too strong, even when lowering the strength). Style can be a bit temperamental, so pretty much anywhere in the range 0.2-1 will work depending on how strong it decides to come on for a specific prompt or LoRA combination. Trained on Anything 4.5. Samples generated with Anything 4.5 and AnyPastel. Also works well with Kenshi. I've also uploaded a TI version which works well sometimes too. I recommend having both for situations where the LoRA might be too much. |