-- A capable multi art style model that can also do 3d --USE DANBOORU PROMPTING, NOT TRAINED ON NATURAL LANGUAGESaikyo is a model that I trained on a big chunk of my digital art datasets that were above 1280px, it is meant to have a wide variety of art styles, and be as creative as possible. It can do both 2d and 3d (90% of the dataset is 2d, so the 3d capabilities are limited, but still decent), it can do good backgrounds and also detailed clothing/armor (one dataset dedicated just for detailed armor and clothing pieces). A lot of the datasets included are from loras that I have published, keep in mind that I didn't merge those loras, but instead trained the model with their datasets + mutliple other datasets that I haven't made loras from.Trigger words from existing loras which should affect the art style: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1759279/samdoesarts-style-wai-nsfw-illustrious-qwen">samdoesarts</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/2121557/the-most-nonchalant-lora-in-existence-or-ill-nai-pony-zit-lora">nonchalant</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/2205091?modelVersionId=2483192">roughline</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/1945988/jcsw-style-wai-nsfw-painterlyimpressionism-style-for-illustrious">jcsw</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/2040908/ya3di-yet-another-3d-illustrious-or-multiple-versions-available">3d, realistic</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/1745914/marvel-rivals-art-style-wai-nsfw-illustrous-lora">r1vals</a>If you don't use a trigger word the model will pick an art style based on the prompt and seed (so mostly random). From my testing the only art style that seems to be a little bit "lost" is "nonchalant", the gens with it look flatter, but the art style isn't fully there, you can always try using danbooru triggers like "flat_color" or "impressionism" etc to nudge the model towards a certain style. Don't be afraid to mix and match the style trigger words ;)You can use pretty much any settings for inference, the workflow is attached to all my images so you can check it out by drag and dropping one of my images into comfyui. (currently using dpmpp_sde/karras with 30+ steps, with the custom sampler nodes)

Saikyo [Illustrious]

Next version will be trained in the cloud where I can use a more advanced setup which should result in even better results overall.

Files with SHA256: 80f17b7d79265cd1d0dc31518cdee673dcebf52f1d83665dc8fca190d91bd739