A Flux LoRA that will make any women in the image more busty.Important: This model is 100% fictitious and is not based on images of any real person. All dataset images originated with images generated from still image checkpoints and that evolved from there using only AI rendering. No images (neither actual nor AI-generated) of any real person were ever involved in the making of this model.PLEASE READ the "about this version" statement on the right side of the screen. There are lots of details about the latest version including prompt tips and information.<s>The model was trained on clothed women only, so this model won't help with Pony-style nude images.</s> (includes nude/topless since version 2.0) It works well with all kinds of outfits from loose and baggy shirts to skimpy bikinis. There is no trigger word needed. Just add the LoRA to your pipeline and... instant cup sizes.The results are mostly in the "large" or "very large" range with occasional results below or above that. This is relative to people in real life, not necessarily to other AI images you may find here on Civitai. The look is usually natural but not always.How to use:<ol><li>Add the LoRA to your image creation</li><li>There is no step 2.</li></ol>Possible issues:<ul><li>Sometimes the women have unusually hard nipples poking through the outfit. I haven't found any way to tweak my prompts to remove that, so if you see it you're stuck with using a different seed or else scale back the LoRA intensity to 0.5 or 0.6.</li></ul>While no trigger word is needed, you can ask for even bigger by specifying "busty", "large breasts", "huge breasts", or "oversized breasts" for more extreme results.For reference, all of the sample images were created with Flux Dev, fp8, on my personal machine using comfyui. Each image has the comfyui workflow embedded, so if you want to re-create it you can download the full size image and import directly into your own comfyui. The workflow I like to use for general purpose Flux generation is the one I published here: <a target="_blank" rel="ugc" href="https://civarchive.com/models/906003/flux-easy-multi-controlnet-selector-workflow-for-comfyui">https://civarchive.com/models/906003/flux-easy-multi-controlnet-selector-workflow-for-comfyui</a>That workflow has a plain Flux processing option so you don't have to bother with controlnets if you don't want to.Enjoy, and please don't forget to post any images you make with this LoRA back to this page. I'd love to see what you can create!

BustyWomen - Flux

There are some incremental improvements in Version 3.0.<ul><li>More face variety, less of the "beautiful woman Flux Face" lady</li><li>Better response to different poses and camera angles with less weird anatomy.</li><li>Slightly bigger breasts without using extra prompts, allowing you to scale back if you want to.</li><li>Profile shots no longer completely unrealistic.</li><li>Unrealistic nipple bump artifacts are much more rare.</li><li>Red-haired and auburn-haired prompts generate more realistic results.</li><li>Used "topless" keyword on every training image with visible nipples. You can use that in your prompts if you want those results.</li><li>Used keywords "cleavage", "sideboob", and "skindentation" where appropriate. You can incorporate those words into your prompts and it will help guide the results. As always with AI images, that isn't guaranteed.</li></ul>I added many more images to the training set bringing the new total to 376 images with more faces and poses included. To create the example images for this version, I ran a random prompt on auto-repeat for hours and made 247 images. Out of those, only 9 had bad anatomy ranging from messed up fingers to extra limbs.I didn't work as hard at the NSFW quality this time, so honestly v2 might be a bit better for that.Prompt tip: I found that hair color prompts will often override other prompts that cue about nationality, race, and skin. My random prompt below doesn't mention hair color for this reason. If you want unusual combinations, try asking, for example, "an asian woman with dyed blonde hair". I tried to use "dyed" in places where appropriate in training photos, which should make that a useful prompt term.Overall, I think this version is more consistently high quality with more variety and flexibility.Random prompt I used to generate the samples:<pre><code>Photo from a __cameraangle-flux__ camera angle. A beautiful __skin__ woman is __action__ in a __cyberdelia/locations/cyber_rooms__. She is wearing __trendy_outfits__. Her hair is __hairlength__ and __hairstyle__. Cinematic high quality photograph.</code></pre>

Files with SHA256: 02a404b607837a310c578d0d3844c6311b5b9f844fa5b470d085d3078a425949