The anima model is very promising. When a proper e621 fine tune is available, I plan to do more with it. Here is a LoRA merge fine tune with as much compute as I can throw at it at the moment. Let's call this a beta. This merge has a good amount of e621 tag knowledge and concepts, but e621 artists are barely present. including the tag 'deviantart' in inference is highly recommended.

anima toony merge

Additional training with natural language captions tagged with JTP Pilot2 Furrence, no caption shuffle and no flip_aug.For improved anatomy, 30-50 steps and euler sampler are recommended (sweet spot seems to be about 38 steps). Natural language will yield the most coherent images. See sample images for prompting style.WEAKLY recognized artist tags:@accelo, @aycee, @bagelcollie, @baphypaws, @blitzpony, @braeburned, @budoti, @butterchalk, @chunie, @cobaltsynapse, @coffeesoda, @cometdragonite, @cytricacid, @dacad, @daftpatriot, @dagasi, @dingbat, @doodlemark, @edjit, @enginetrap, @fantine, @funkybun, @hinar miler, @hioshiru, @indexer, @jumpy jackal, @kashmere, @keffotin, @kenket, @keovi, @k1ngl30n, @letodoesart, @loimu, @mawmain, @meesh, @orlandofox, @reign-2004, @ruaidri, @sushiamazing, @tzarvolver, @waspsalad, @zackary911, @zaush

Files with SHA256: 5198b57b34b1ec6ee9556453d24dce7c8b89ab86b2164582f0b3b9333a9049b9