If you like what I do, buy me a coffee~ https://ko-fi.com/budace
I also started a Telegram channel where I share random generations that I make: https://t.me/InflaitableWonderland
4.0 Update
I lost the previous dataset so I had to rebuild this one from scratch. I tried to more evenly spread out each concept as I did. The quest toward perfection continues. As always, you can mix and match each concept tag to your heart's desire. This dataset has a lot more monorump and head inflation support if you're into those things.
Still has some consistency issues and still has trouble with the different size tags. I chalk that up to probably bad tagging. I'll fix it in the next version, I've been working on this since before Christmas lol.
3.0 UPDATE
Still follows similar tagging language as my other inflation model here - https://civarchive.com/models/210476/inflation-lora
Of course omitting female specific tags like breast expansion and using more male specific tags like penis expansion.
Tried a completely different training technique with this one. I'm pretty happy with out it turned out.
_________________________________________________________________________________________________
Model 1.5 update:
Trained for 50 more epochs and it's still slightly lacking in some aspects, but overall I think it improved the shapes a lot. Also this model works really well with AWP Mix.
XL version of my other male inflation lora you can find here - https://civarchive.com/models/336297/male-inflation-lora
Same principles and tags apply here, I like consistency and ease of use XD. For more extreme shapes you would need the same tags as my other inflation lora found here https://civarchive.com/models/210476/inflation-lora
Things unique to this would be tags like blimpdick, or blimpballs for when hyper isn't big enough. Also muscle growth and pecs expansion, for those specific areas. Also if you if you like your dude stretching clothes there's bulge expansion, hyper bulge, and blimpbulge. Each one is intended to be used by itself as a sort of size selection. using something like bulge expansion and hyper bulge together will basically just make the bulge hyper as if you used that on its own anyway.
There's also muscle growth and hyper muscles. As well as pecs expansion and hyper pecs.
Finally with this training I added sitting on balls as a tag for some more fun posing options with huge balls~
Description
FAQ
Comments (9)
Does it worked with non furry?
It should work fine with any Pony based XL model
IME, this series has always worked decently well with non-furry characters. however, using many tags in parallel, or more extreme/niche tags like "spherical inflation" and especially "living inflatable" tend to result in furry features being added. reducing the LoRA strength helps but unless more human data is added to those tags you sorta have to pick between compliance and feature leak, at least right now.
there's LoRAs that can help with doing non-anthro living inflatables if you need that, and they tend to interoperate well enough.
@baph0met_balloon Yeah it's not easy to find male inflatable images for training data, even less so with non furry
@budace there's some about, you might try deviantart or pixiv. of course then you'd have to do all the tags yourself
Next time you do comparison grids like that, can you use the same seed for each image? It's hard to compare when all of the images are wildly different.
I'm open to being corrected if I'm wrong, but I'm not sure those are different seeds. The poses are very closely matched; what's different is the style, which could easily be the effect of the model's dataset being strongly based on illustrations showing through.
Also bear in mind that the stronger the levels of inflation, the more 2d illustration-y it'd look due to such sources essentially needing to be 2D illustrations to depict those sizes.
@LlunaBlue You are pretty much correct. This lora isn't really designed to be a sequence maker. The vast majority of the time each size tag is from different artists with different styles and different characters. I have tried my best to make the different styles in the dataset irrelevant without getting too bogged down in tags. Also 99% of my dataset is illustrations, especially the male set. It's not easy to find photorealistic inflation of males in the sizes my lora makes. Also doesn't help that I'm training in some wild concepts that no checkpoint really handles all that well. You probably could get more consistent character style with more character details in the prompts. I just don't add that to my examples because I want it to remain clear that my lora is really only handling the shapes, not necessarily the character.
@budace I'm fully on board with your process and the sheer amount of effort you're putting into these, good sir! :)
(I'm also feeling a little bit of an ego boost from my insights turning out to be correct, hahaha)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.