Trained on Hassan's Anything model. From my experimenting, this model is a little bit rigid/overtrained, but it's almost by necessity because of how easily the anatomy can go wrong. But I'd say it delivers a high quality image in txt2img about 60-80% of the time with a decent prompt. I might revisit it later with a lower learning rate and adjusted dataset rather than train the current one any further. As with most models, the hands are hit and miss, so prompts that give an explicit gesture or outright conceal them like "folded arms" or "hands behind head" are recommended if you are struggling to get good ones.
(There's also some Ranni the Witch in there, as she makes up a lot of the best quality "four arm" art out there, so if you don't want her explicitly, you might want to add blue skin and witch hat to your negatives.)
Trained on Hassan's Anything model. From my experimenting, this model is a little bit rigid/overtrained, but it's almost by necessity because of how easily the anatomy can go wrong. But I'd say it delivers a high quality image in txt2img about 60-80% of the time with a decent prompt. I might revisit it later with a lower learning rate and adjusted dataset rather than train the current one any further. As with most models, the hands are hit and miss, so prompts that give an explicit gesture or outright conceal them like "folded arms" or "hands behind head" are recommended if you are struggling to get good ones.
(There's also some Ranni the Witch in there, as she makes up a lot of the best quality "four arm" art out there, so if you don't want her explicitly, you might want to add blue skin and witch hat to your negatives.)