this makes stylized thicc furry women

the model was trained on a dataset of my favorite images on e6, which is approximately 5K images.
tested with Yiffymix V34.
the model responds well to flat styles/artists such as Cumbread, Haps, Claweddrip, Inu-sama, Buta99, and Securipun.

I usually use all of those artists in my prompt because I like how it combines them.

I recently replaced the v2 with an uncaptioned version because I discovered some contrast issues with the captioned model, I will be attempting to remedy that issue, however, currently, only the uncaptioned models appear to function as intended.

I hate it when my denoising model adds noise to the image, so I trained the model to include the tag "jpg compression" for all images that are jpgs this means you can remove jpg compression artifacts from generating by adding that tag to the negative!

the model was trained at a resolution of 1088x1088 so it performs better with higher resolutions.

what is the difference between vpred and non-vpred/fluffyrock?
the vpred models were trained on the fluffyrock vpred base model, so they should work with vpred models the best. if you don't have a vpred base model, then I would recommend using v2.0

what is the difference between captioned and uncaptioned?
Captioned models are trained with captions attached to the training images, which means that they will have certain words that the model might recognize as a "trigger word," and are generally more controllable. On the other hand, uncaptioned models work more like sliders, where there is no activation word. If you increase the weight of the LoRA, it will exhibit more of the training data (in this case, larger ladies). However, the downside with uncaptioned models is that they won’t have a good way to control the output, so it might make a pregnant-looking woman and even if you add “pregnant” to the negatives, it might not change much.

I made this LoCON model to simultaneously learn how to make LoRAs/LoCONs and also make a model that understands my favorite features better.

Donations

Training models is expensive, if you like what I do, consider supporting the development here!

https://ko-fi.com/yolup