<h3 id="undertale-loras-for-sdxlflux-mqq5mgy00">Undertale LORAs for SDXL/Flux</h3>I've noticed a lack of Undertale LORAs for SDXL, so I've trained one on Animagine XL v3.x. It can work on some other models as well. I also trained a PonyXL and Base model version s and right now Flux LoRA in process.Training Data for SDXL:<ul><li>Used around 800 images in total, with some later additions</li><li>Can influence style quite significantly (mostly 3.0 models)</li></ul>Flux LoRA:<ul><li>Currently being trained using split mode and fp8 version of dev model</li><li>Quality uncertain at this time</li><li>Uses smaller dataset of 558 images and less repeats, Florence2/JoyCaption captioning + corrections by me</li><li>Possibly would retrain at some point</li></ul>Characters: Asriel, Chara, Frisk, Muffet, Sans, Papyrus, Undyne, Gaster, Toriel, Mettaton, Grillby, Alphys, Flowey, Asgore, NapstablookNote: Quality varies between characters. In version 1.0 of Animagine, avoid generating Asgore or Napstablook due to overfitting. Version 1.1 and onwards work better, but caution is still advised.Parameters:<ul><li>Trigger words: character_(undertale)<ul><li>Some characters work well with trigger words alone</li><li>Others require additional prompting</li></ul></li><li>Flux: Natural language prompting</li><li>SDXL: Booru tagging </li><li>The _(undertale) part isn't strictly necessary</li></ul>Model-Specific Notes:<ol><li>Flux:<ul><li>Use guidance of 3.5 or higher for better character likeness</li><li>Higher guidance may increase stylistic differences between images</li></ul></li><li>Animagine 3.1:<ul><li>Model 1.2 surpasses previous versions in flexibility and capabilities</li><li>Stylistic differences between versions are subjective</li><li>Personal preference: Model 1.2.5 for style, but has biases due to initial overfitting</li></ul></li><li>Base XL:<ul><li>Second model trained on base model for broader compatibility, as a substitute when other models don't work</li><li>Intended for use with any model, but struggles with anime models</li><li>Examples not limited to Animagine v3</li><li>Currently not fully accurate, performance varies by model</li></ul></li><li>PonyXL:<ul><li>Trained on civitai</li><li>Can produce good outputs</li><li>Missed some character features</li></ul></li></ol>Note: There may be stylistic inconsistencies, so it's better to specify your preferences when generating images.

Undertale [XL, Pony, Flux] LoRA

Asgore and Napstablook overfitting problems are mostly fixed, now it is possible to generate variations of them, even if still a bit rigid After additional training, other characters are generated with a more accurate resemblance, practically complete Style is more consistent Easier to create images with more than one character Have to note that model still can generate images that are not fully connected to the prompt

Files with SHA256: 17b6ccf9188ee9260cd697f0bb643c2e1482106eddf3a389b8a677e6e64fe961