Overview
This model was a commission.
A helper model to create more muscular women, while remaining cute and feminine.
The trigger is muscular as well as describing the size and shape of her muscles.
This model cannot create NSFW on its own, but works with LoRAs like SNOFS (for Flux/Krea 2) or detailers on lower strength.
Krea 2 Turbo - V1
Use the WAN 2.1 VAE for more realistic outcomes. Use the filter bypass LoRA for better results.
Lower strengths give a more buff/bulky look, while higher strengths give a vascular/bodybuilding look.
Training model: Krea 2 Turbo + Training Adapter
LoRA strength: 0.3-1.0
Sampler: euler / euler a / euler+beta57
Steps: 8+
VAE: WAN 2.1
A photo of a extremely muscular woman. Her incredible bulk commanding presence in the gym. She is flexing one massive peaking bicep, looking into the camera with a smug smile. She has long blonde hair, blue eyes and a cute young face.Flux.2 [klein] 9B - V1
Training model: Flux.2 [klein] Base 9B
LoRA strength: 0.5-1.0
Sampler: euler / res_2s / res_multistep
Steps: 4-12
CFG: 1.0
A photograph of a muscular woman with an elite bodybuilding physique is flexing her enormous muscular arms. She's wearing a pink tank top and shorts. She looks at the camera with clenched teeth.ZIT - V1
Not compatible with Z Image Base.
Training model: Z Image De-Turbo
LoRA strength: 0.5-1.0
Sampler: euler / res_2s / res_multistep
Steps: 8+
CFG: 1.0
Prompt example:
A photograph of a muscular woman sitting on a park bench. She's lifting her t-shirt by the hem to reveal her extremely defined abs. She's has a cute young face, contrasting with her enormous physique.Description
FAQ
Comments (5)
This is a impressive LoRA. Very well made.
The Z version can be juiced up even more with slider LoRAs without any degraded quality.
And the fact thats its trained on the de-turbo is even more impressive to me. But why not on base, if i may ask?
@Loraholic because De-Turbo is consistently better for training than Z-Image base. They have different internal representations due to Turbo's distillation method. Using a LoRA that was trained on base can quickly look overbaked.
@cupra I have the same experience with concept sliders. But for outfits, character etc i have had much more luck with base. But i guess it also boils down to the dataset and settings as well. You have clearly found a golden setup here.
Sliders are miles better on Base tho. All my sliders are made on base. But that is also a completely different type of training.
Thanks for the answer.
@Loraholic I tried base a lot, but for all concept work, especially large scale anatomy like macromastia or giantess, it didn't translate at all into turbo. Maybe that's the difference, it works good on the lower noise, but not on the higher noise.