Type | |
Stats | 490 3,813 |
Reviews | (96) |
Published | Nov 20, 2023 |
Base Model | |
Training | Steps: 2,600 Epochs: 10 |
Usage Tips | Clip Skip: 1 Strength: 1 |
Trigger Words | Machamp |
Hash | AutoV2 D37472C462 |
Machamp LoRA trained on NovelAI with Kohya training scripts using the DAdaptAdam optimizer at 512x512 resolution. Very difficult to work with because of the extra arms. Most models will turn the additional arms into something else, in YiffyMix's case that was tails and wings. Your mileage will vary depending on the base model you're using. Adding (tail:1.4) and (ears) to my negative helped improve consistency. Arm count can vary wildly depending on the prompt, with clothing weakening the LoRA's strength significantly. I recommend increasing the LoRA strength when prompting for clothing. Added (belt) as a trigger word, prompt for that to get Machamp's belt. Despite all of the issues, great results are possible. Should have wide compatibility with model mixes based on NovelAI, which is most of them. Trained on a mixed male/female dataset.