This is a LoRA for SDXL focusing on wheelchairs and people using them. Actually, it is an output from one of my abandoned projects. Please don't expect too much :) and handle with care.Please feel free to improve it if you are interested in it. (Or just ignore this and grab another.)<h2 id="heading-243">Features of the training dataset</h2>It consisted of:<ul><li>about 200 photos of one person who is on their wheelchair, containing less than 30 mature shots (nudes or underwears, no porns),</li><li>about 50 photos of groups of people with one, some, or all of them in wheelcharis, and</li><li>about 20 photos of an unattended wheelchair.</li></ul>It contained usual traditional wheelcharis only. That is:<ul><li>specialized wheelchairs, such as for basketball or racing, were not included.</li><li>about 90% were manual wheelchairs and 10% were manual-electric hybrids with two large wheels. </li><li>Motorized wheelchairs with small driving wheels were not included.</li></ul>All samples were photos. No illustrations, 3D graphics, or anime were included.<h2 id="heading-244">Known issues</h2><ul><li>Wheelchairs often look strange. I believe it's better than the vanilla SDXL 1.0, but I consider it is still far from acceptable. You have been warned.</li><li>I have a feeling (though not sure on this point) that varieties of human this model can infer are too limited. If you generate 10 pictures by a same prompt, for example, you will often find two images showing a very similar person. Perhaps an overfitting.</li><li>... and many more.</li></ul><h2 id="heading-245">Acknowledgment</h2>@asantiagowow655: Thank you for pushing me to post this model.

Files with SHA256: c2e6ce4c46a38c6a49a0def2a296024eeaf1608bcb2cc662a71540cff20b1bf3