Who?IllustriousAs ever, trust this model to get good training results. Generally speaking, it's quite good at depicting him topless, nude, and so on.AnimaSet the weight to 1.3 to get Regulumon generating with accuracy, albeit at the slight expense of versatility. Weight 1.4 and above brings diminishing returns and also causes concept bleeding into other characters. It's possible I've underfitted this LoRA during training but that's hard to believe given where the convergence typically happens. Civitai's AI Toolkit trainer likewise understood all the details but overfitted it quite a lot (which is why I opted not to use that version).I've updated the showcase with new images to demonstrate the difference with increasing the LoRA weights, and will replace some of the others over time.<s>Write</s><s> furry male</s><s>, </s><s>beast man</s><s> or</s><s> Digimon </s><s>when prompting since the LoRA is prone to mistakingly giving Regulumon a human face. On that note, this has ended up being a harder LoRA to train due to the decisions I made, as its the first time the AI hasn't understood it's looking at a Digimon-looking character despite a lot of similarities to Leomon, which I had banked on. His completely absurd proportions has resulted in a lot of ambiguity for the AI. On top of that, that art style has been baked in, and is harder to change. Lower the weight of either the trigger word or LoRA down to 0.75 to help it with this.</s>

Regulumon

<ul><li>Main features are here</li><li>Can't do nudity very well</li><li>Stubborn art style</li></ul>

Files with SHA256: 853c3e68f306d84b5baa268b9015ca47af2918b68ad5585837a92dddce3deee4