https://civitai.com/models/210476/inflation-lora LoRA weight range: 0.8-1.0 Author notes, budace: UPDATE 2

Rolling out the 3.0 versions trained against some different checkpoints. The checkpoints used in training is listed in the name. 3.0 added a ton more images to the dataset and I reworked some more tags. Now it works better if you add inflating in your prompt. same overall concepts as before but now spherical inflation works best if you add puffed arms and puffed legs. I'll add more to this update as I roll out the other models

UPDATE

Updated the dataset with more images and cleaned up the tags a bit

I have been testing my dataset against different checkpoints and honestly I can't decide which one I like best. These are all the exact same dataset and tags trained against different checkpoints or with different optimizers. Most models I run with the default adamw8bit optimizer in the script but I am trying to bring a training farm computer into the mix and it can't handle that hence the Prodigy experiment. But yeah with each version I'll put the checkpoint name and optimizer if it's anything OTHER THAN ADAMW8BIT. Each one does different things a little better than the other versions so I can't pick a favorite.

Definitely still needs a lot of work on concepts like -spherical inflation- Usage notes, tagging and misc.: Since it's trained on FluffyRock it works best with e621 tags.

Use "expansion" for some general enhanced curves. IE breast expansion, butt expansion, belly expansion.

Use "hyper" for more extreme curves. IE hyper breasts, hyper ass, hyper belly.

You can mix and match any of these as much as you want.

"Living Inflatable" makes for some balloony or pool toy looking fun and can be further played with by adding sex doll.

Add "Floating" for the people who enjoy the floaty end of the inflation spectrum

Finally there's "spherical inflation" for that full rounded shape. This one still gives me trouble and generally only works best on checkpoints that use E621 tags so far.

I've been working on this model for quite a while. Probably being overly ambitious but I wanted to make an inflation/expansion model that could cover just about every concept. This model was trained on FluffyRock checkpoints so it definitely works best on other furry models but it does pretty well on non furry ones too but some weirdness might come through. With the weight at 1 the shapes really come through but it can overpower the style of the checkpoint you're using. Generally I have found the sweet spot to be around 0.8.

Anyway I think that's enough info, I hope you enjoy~