The model's intended purpose is to create images of Telma that match as closely as possible to the source material. The training dataset contains NSFW images and is freely accessible. Suggested Settings: The model currently does far better with a CFG between 1.0 and 2.0. I also found that between 8 and 16 steps with the SDXL 8-Step Lightning LoRA work well. It seems like the eyes are a bit better defined when you move closer towards 16 steps, which is odd, considering you're generally supposed to match your steps to whichever lightning version you're using... NOTE: This is my first attempt at training a LoRA and is thus very much a "Work in Progress." The current model has not undergone any sort of fine-tuning and thus has a few kinks that need to be worked out. The main issues are baked in clothing and poses when using the "3D" and "Video Game Render" tags, which were used to indicate when the original style is desired. The nude body for Telma also needs a great deal of refinement in order to produce clearer, cleaner images. It also has an odd tendency to create images with two Telmas in it. I'm currently working on refining the dataset, upscaling and cropping images, and creating additional high quality training images with SDXL via a simple "txt2img -&gt; inpaint -&gt; photoshop cleanup" pipeline.<a target="_blank" rel="ugc" href="http://Flux1.Dev">Flux1.Dev</a>: I'd like to try training a <a target="_blank" rel="ugc" href="http://Flux1.Dev">Flux1.Dev</a> version of the model once the dataset and tags have been refined a bit more, but my hardware wouldn't be able to handle training or image generation for it, so I would likely need to either purchase or be gifted buzz to do so. A <a target="_blank" rel="ugc" href="http://Flux1.Dev">Flux1.Dev</a> version of the model would likely be able to produce better quality images that I could then use to further refine the SDXL model, which is my primary goal. I would also share the <a target="_blank" rel="ugc" href="http://Flux1.Dev">Flux1.Dev</a> version assuming it wasn't a complete failure.Any Buzz gifted will go strictly towards LoRA training to improve the model.For those of you with experience training character LoRAs, any guidance/suggestions/feedback you can offer is GREATLY appreciated. Thanks!-- Ballz2DaWall

Telma

Model was trained using the following tags for images of the 3D model and screenshots from the game: 3D, Video Game Render

Files with SHA256: 605708036f62b5ac2019caccc35cad03795e551c81a3af3db1cd080475567f56