This DoRA is trained on a 1.7k images of varying belly shapes and belly clothing states for anima and others. I felt like it was better to combine and improve my existing belly shape LoRAs into one for ease of use and training.
Update 2026/07/19:
Trained on Anima, and added two new shapes.
long belly: similar to torpedo, but is typically much larger and longer overall.
sprawling belly: The belly is sprawling across an object, with an overall flattened and wide shape. It helps to describe the object it is sprawling over in the prompt, like floor, table, etc.
Important tags (same as Hyperfusion tags):
torpedo belly (fairly pointy and torpedo shaped)
sagging belly
distended belly (when the sides or ribs are being stretched by belly)
spherical belly
stuffed belly (stuffed with food or liquid, tends to be a more vertical shaped belly)
"full stomach" is a helper tag that describes the top part of the stomach being larger than the bottom for stuffed bellies.
wide belly (only works with straight-on poses)
long belly (longer than torpedo, but does not have to be pointy)
sprawling belly (typically large flabby bellies sprawling across a surface)
to a lesser extent:
low navel (navel is placed low on the belly)
hanging belly (unlike sagging, the character must be bent over in some way)
exposed belly
clothed belly
innie navel
outie navel
prey shaped belly (didn't learn well)
Probably others I haven't tested yet from the hyperfusion tags list
belly_sizes:
medium belly
big belly
huge belly
gigantic belly
hyper belly
hyper pregnancy
obviously some belly tags the model already knows were also included:
pregnant belly
bloated belly
chubby belly
etc...
Description
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:
beta2=0.9999, soft_min_snr=2, no weight_decay, decouple=False
weight decay was causing issues with DoRA. The model came out better without it.
FAQ
Comments (7)
I'm getting the same issue with baked/burnt colors and crazy artifacting when trying to use this through Comfy, even with a setup that works local to CivitAI's generator. Would love to see a cleaned-up version for newer Illustrious mixes.
Its only going to work with noob_vpred and decedents of that. No guarantee for non vpred illustrious finetunes. If your still having issues on noob_vpred, then its probably a config issue.
Something is wrong with this model. For instance, trying to remix an old image that looks good will often end up with a terribly colored image even if no changes were made to the prompt.
Genuine question, is there any hope for this to be made into a illustrious model in the coming months or so?
I would love an update, because the torpedo shape would be extremely useful to have in anima. Its good with large bellies, but it really cant do that shape on its own.
ok, when one of my gpus gets some free time, ill see what I can do
@throwawayjm The good news is, training with anima is super fast. I've done a few test ones with it, and it only takes a couple hours . That's on an 8gb card,
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.
