Flat chest, small breasts, very small breasts, very flat chested, flux, nsfw, adult
Trained only on the chest. Use with any face, ethnicity or body type.
Trigger "flat chested"
Lower strengths increase breast size a little. For as small as possible use 1 or higher strength.
Description
FAQ
Comments (12)
Sorry, I can't figure out how to add text/instructions. Just use "flat chested" as a trigger, and you can lower the strength to increase breast size. Strength 1 or more for the smallest.
Open your lora page and next to Details (top right, right above Type) , select Edit and then Edit Version Details. That will allow you to change your trigger and other details =D
I also recommend you avoid using spacing for triggers as Flux will read each word individually and can get confused. flat-chested, flat_chested, or flatchested would work better. That being said, Flux can still ignore that and see those are general terms to apply in a prompt. Triggers are best when very unique and something Flux cannot misunderstand. Fl4tch3st3d, for example cannot be confused. That would need to be trained into every caption though and specified as a trigger during lora training.
Very unique triggers seem to help a lot =D
@wodz30 Thanks!
@wodz30 Thanks. My first two Lora's are okay, but I did them incorrectly based on my experience with another AI. Each one has the breasts of multiple women, around 5, and what we get is an average, when my intention was variety. On the "small and saggy" I used a single model, which I think is more appropriate for the number of pictures I'm allowed, and it came out much better. I may come back to these first two at some point.
@ZellArista If you look around at othr breast loras for flux - most are pretty crappy tbh. Only some of the samples look decent and then actually using the loras tend to yield very poor results.
I am generating 1,000 images right now (local generation) using your flat chest lora and the results are fantastic with good regularity. Zero issues.
For good variation, I would recommend 10 sample images per breast model. Trigger each size with a different caption as well and use something unique like sm411breasts for the 10 samples, and so on. Once your data set and captions are completed, you get into your lora training config and specify each of the trigger words. If you end up with 50 images in your dataset (as an example), you should look to push 5,000 steps. Many of us are using ai-toolkit for flux training. You can even do that online with RunPod (have not used it) - https://github.com/ostris/ai-toolkit/
@wodz30 Thanks for info. I'm limited on what I can do on my own system because I'm using a laptop with an RTX 2080 in it. I'm looking to possibly build a desktop when the new RTX 5090 becomes available next year.
I have a lora on tuberous breasts/puffy nipples coming in the next day or two if you are interested. Suggestions welcome
Your lora is one of the very few here that are totally not changing the face. It's great! If you train without the head, what's your process? Do you just crop the head out? Or do you use transparency? Or a mask? How do you do this? I am using fluxgym but can't find a way to do this. Can you help?
I cropped out everything irrelevant and just had the upper torso. No bits of hair or anything. For boobs, as long as the AI can figure out what it's looking at, you're good.
@ZellArista very interesting. I was wondering if there was a way to "mask" the head so that it knows there is a head (to help it figure out the rest) yet does not take it into account in the lora. But I don't know how to do this exactly... thank you.
@sparkleboy599 There's no reason to let it know there is a head. It already knows this.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.