V2: Better image curation and local training with LoRA Dataset Studio on my venerable 3090. Slight improvement all around, though all the caveats about how different checkpoints respond (VERY) differently and scheduler/sampler combos still apply. Again, I'm offering a couple of alternatives, both higher- and lower-trained, as the trend of really low step counts turning out some bangers continued with this version. ----- Adds a reasonably large blue fox fur coat to your subject. The training data had some issues, but it seemed to turn out well, so here we are. For the sake of specificity, it was trained on images of women wearing blue fox fur coats, but it may work otherwise. I have no plans to find out.Usage Tips:<ul><li>The trigger is "bluefoxfur," but it's going to kick in with any "fur coat" prompt.</li><li>Sampler/Scheduler: Notably dependent on the checkpoint. Start with whatever the model/checkpoint recommends and go from there. I would recommend using either a “detail-oriented” scheduler or sampler, not both; texture starts to get a little overbaked otherwise.</li><li>Example: ER-EDE or EXP Huen 2 x0 SDE pairs with Simple. Or Euler with Beta_57.</li><li>I get good results with Euler Ancestral and Beta_57.</li><li>For basic “fashion” shots, 1 is fine. If you plan to have a lot of body parts interacting, you may want to dial it down a bit.</li><li>At a lower strength, it can be a good ‘texture booster’ to make the model’s fur look a little better without strongly overriding what it would have given you anyway.</li><li>The coat will be “large” by default, but feel free to go nuts on your favorite adjectives: huge, oversized, voluminous, etc. It should respond favorably.</li></ul>

Blue Fox Fur Coat

Lessons learned applied. Better training set, primarily focused on "full pelt" coats (the kind with fewer vertical seams, basically), and fewer 'distractions' in the image.

Files with SHA256: 404bf7b1bee8b258675de66ffff377ae1981fb95f9e215cbbb9843245ad9f43a