The latest version of the model was trained on the new Z-img base model and will work on both the Z Image Base & Turbo versions. BUT to get it to work on Turbo you will probably need to increase the strength to 2.0, at 1.0 it was having almost zero effect for some reason.
Made as a user request. If you don't know what this is, Urban Dictionary is here to help: http://torpedo-tits.urbanup.com/16189773
A LoRA trained to generate woman with Torpedo Tits / Conical Breasts. This model was trained on a curated dataset of 136 images with light/minimal captioning for 3000 steps. Generates a decent pair true to the concept most of the time and seems to work across different sizes and nipple types.
Main Trigger Word:
a woman torpedo shaped breastsAdditional Trained Concepts
puffy nipples
close up shot
front view
low angle shot
side profile
three-quarter view
See example images for prompt examples.
What it's capable of:
Multi-Style: Works well for photo style gens and surprisingly very well for Anime considering there was none in the training set but you may need to turn down the strength to .7-8 for anime generations.
Consistent Framing: Generates the concept well at most angles of framing
Archetype & Setting Variety: Can generate a wide range of female archetypes and integrate them into diverse settings/environments
Body Diversity: Responds very well to descriptors for age, body shape, and unique features.
Description
This version of the model was trained on the new Z-img base model and will work on both the Z Image Base & Turbo versions. BUT to get it to work on turbo you will probably need to increase the strength to 2.0, at 1.0 it was having almost zero effect for some reason.
FAQ
Comments (5)
Sup bro, is there any difference training on Z-Base instead of Z-Turbo? I mean, in the final quality of the pic. Or the final result is the same?
Also, you use SeedVR2 to get those crispy photos correct?
Training on base doesn't seem any different so far except it does take more time, vram and steps to train.
Since the training works on both base and turbo you get two for one and I think the training on base improves the quality of the lora on turbo. So far I'm actually preferring the output of Turbo using a base trained Lora vs using base and the Lora.
For the quality, the upscale I usually use is RealESRGAN_x2, I have SeedVR2 in my flow but I don't often use it because it's honestly way to slow for my liking when generating 100's of images to test out the Lora. So all the examples I post are uses RealESRGAN_x2.
This but for klein? please
Sorry I don't use klein currently so no training plans at this point, but if that changes I'll post one.
posted a video, thank you, love these kind of tits. I didnt have any issues with other loras at the same time, so can only assume where people complain, its their worklow thats the issue.