This LoRA generates bubble butts for a number of SDXL/ZIT models. Its nearly the same dataset as my SD1x bubble butt but with a handful of larger sizes added, and better tagging.
Use the tag "bubble butt", with (optionally) one of the following size tags "medium ass", "big ass", "huge ass", "gigantic ass", "hyper butt" along with similar Hyperfusion tags.
If you want front facing images try "from front", or "ass visible from front" and negative "from behind, looking back, ass focus" at varying weights.
v4 ZIT Changelog:
Couldn't resist training a lora on Z-Image-turbo. Went pretty well, so ill probably be training more when the base model drops.
This one is just a quicky, probably not fully trained, but good enough as you can see.
Dataset is fully anime/furry, so you need to include realism related tags and words if you want that.
v4 WAI_IL Changelog:
No changes, just re-trained on WAI_IL-v14 because I was in a tinkering mood.
v4 noobai_vpred Changelog:
Not much changed. Added a few more images/captions/tags, and trained on noobAI_vpred_1.0 with somewhat similar settings to v3
Same tags as v3, you already know what to expect
"hyper butt" has a greater effect on this noob_vpred version as well if you want it really large.
Description
No dataset changes between this one and the last (hence still v4). Just wanted to try out ScheduleFree training for once, and went back to using LoRA_LoCon instead of DoRA for a change of pace. It went well enough (but DoRA still turns out slightly better for me in general).
Really just wanted to see if WAI_IL was any harder to train on than noobAI_vpred, but it was pretty much the same.
See last version for training config.
FAQ
Comments (9)
The thighs is too slim in new version
You can blame WAI_IL. NoobAI had plenty of furry content which probably made thicker thighs overall. I used the same images as my last version.
There were 150 images labeled "bottomheavy" in this dataset, you can try and see if it helps. Otherwise use any "thick thigh" related tags.
AWESOME!
Dear throwawayjm, your lora is the most used for me. It's my favorite. I recently tried a DORA, and I found it to be incredibly more powerful than a familiar Lora or Lycoris. I have to reduce its strength to 0.4 at most because it's ultra-precise. Are you interested in making a DORA version for Noobai V-Pred 1.0 version? I'm looking forward to trying it.
The noobai_vpred version is already a DoRA, only this WAI_IL version is a normal LoRA_LoCon. Civitai only allows me to select a single option (which is unfortunate).
I guess they never expected people to use both in the same series :/
One of my favorites!
Gyatt, and I cannot stress this enough, dayum.
@throwawayjm I've a doubt: ZTurbo = FP8?
Not 100% sure what the ask is, but weights are trained and saved as bf16, and as far as quantization during training goes, these were my values (notice only TE quant options were uncommented)
# quantize: true # run 8bit mixed precision
# qtype: "qfloat8"
quantize_te: true
qtype_te: "qfloat8"
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.