I trained this model using a dataset composed almost entirely of real-world photography. My goal was to achieve realistic textures and lighting, moving away from the "plastic" look of purely synthetic datasets. While this implies the model isn't technically "perfect" yet, it does a crazy good job at what it was designed for.
✨ Key Features & Usage:
No Hard Trigger Word: You generally don't need a specific trigger word to get results. However, if you want to emphasize the size, simply adding keywords like huge or big will push the model to the limit.
No Body Type Bias: Almost anytime traning a model like this using real data, most of the models should have a natural tendency to generate "chunky" or curvy women, this model besides shouldn't do it. If it does and you prefer a slimmer physique, simply add skinny to your prompt, and it should correct the body type without issues.
Wardrobe Malfunctions: Be aware: if your prompt includes "cleavage," the model has a high probability of generating an areola slip. If this isn't what you're looking for, try rolling a different seed (though personally, I consider it a feature, not a bug 👀).
⚠️ Current Limitations:
Mixed Sizes: I haven't managed to get the model to generate multiple subjects with different sizes in the same image (e.g., one small, one huge) just yet. It tends to average them out.
Future Updates: This is an initial version. I might release a fine-tuned v2 to fix the mixed-size issue, but honestly, I'm pretty lazy, so enjoy this version for now!
🧡 Consider supporting:
Description
significally increased the size, bringed back the details on the skin, now the rest of the body will appears even less chubby leaving more freedom to the prompt (obviously if you want you can just add chubby to the prompt). added more nude body to the dataset but still can't get the "cat" appear. [if you get some artifacts lower the strenght but it should work great at 1.0]
FAQ
Comments (5)
Since you mentioned that you are using real photos as your dataset, I was wondering why the areolas and overall breasts skin looks like it is enlarged using 2d bulge/mesh warp deformation that stretches the existing skin texture to be larger than the original.
Any chance that your real photo dataset has been manipulated with mesh warp or similar pixel deformation tool to enlarge the real breast photo to be bigger than what they really are and using that as the training dataset? Just curious, as from my observation of your samples and my own limited tests, the result consistently has the stretched skin texture artifacts. Just curious, thanks for the great work.
First off, great model. I haven't tried v3 yet, but judging by the preview images, it kinda looks like a downgrade from v2.75 ngl
hope you'll give it a chance!
@flusk ok so I've been using it for a few days now, and v3 is definitely an improvement, especially on clothed shots, since it doesn't generate excessive cleavage unprompted. I use it at a strength of around 0.3 to 0.6.
Honestly, I dont like this particular lora bc breasts look similar every time and has its ai gloss and plasticness