This DoRA is trained on a 1.5k images of varying belly shapes and belly clothing states for noob_vpred. I felt like it was better to combine and improve my existing belly shape LoRAs into one for ease of use and training. Not to mention noob_vpred training results are just better than pony.Update 2025/02/17:<ul><li>forgot to upload v4 weeks ago, but this new version is just better at all shapes. Use the same tags as v3.</li></ul>This was really just a test of a new optimizer ADOPT mixed with learned_timestep_loss_weighting, trying to find the best combination to make it work. There is still room for improvement, but this was the best of 1.5 weeks worth of test runs, lol Important tags (same as Hyperfusion tags):<ul><li>torpedo belly</li><li>sagging belly</li><li>distended belly (when the sides or ribs are being stretched by belly)</li><li>spherical belly</li><li>stuffed belly (the top half of the belly is same or larger than the bottom)</li><li>wide belly (only works with straight-on poses)</li><li>to a lesser extent:<ul><li>low navel (navel is placed low on the belly)</li><li>hanging belly (unlike sagging, the character must be bent over in some way)</li><li>exposed belly</li><li>clothed belly</li><li>innie navel</li><li>outie navel</li><li>prey shaped belly (didn't learn well)</li><li>Probably others I haven't tested yet from the hyperfusion tags list</li></ul></li><li>belly_sizes:<ul><li>medium belly</li><li>big belly</li><li>huge belly</li><li>gigantic belly</li><li>hyper belly</li><li>hyper pregnancy</li></ul></li><li>obviously some belly tags the model already knows were also included:<ul><li>pregnant belly</li><li>bloated belly</li><li>chubby belly</li><li>etc...</li></ul></li></ul>

belly_shape-sdxl

Same dataset as v3 but slightly different config. Finally had time to test out many different ADOPT optimizer configs, and this was the best result. Only changes from v3 are:<ul><li>beta2=0.9999, soft_min_snr=2, no weight_decay, decouple=False<ul><li>weight decay was causing issues with DoRA. The model came out better without it. </li></ul></li></ul>

Files with SHA256: ef7b606caa0b5e4d1887e2a94968a6157688afc502f8ef107cd9630278f5b305