This LoRA is designed to help produce men with more "average" looking bodies. Flux.1 is notoriously known for producing attractive guys with only six packs/sub 10% body fat. This will help produce more natural looking body shapes without compromising good looking faces.

What version is best for me?

v1.0 Regular - this model has lower epochs and works better for most cases, although the effect is a little less prominent. Produces results with probably around 12 - 20% body fat.

v1.0 HD (High Detail/Higher Epoch) - this works better for a strong effect, but more details from the training data will bleed into your images, altering them a bit more. Produces results with probably around 15 - 25% body fat.

Additional Info

It has been tested primarily with the euler sampler with a simple or beta scheduler

Works well with flux.1 Dev ultra real fine tune and flux.1 dev regular

LoRA strength should be between 0.4 - 1.5. The "intensity" of the effect is proportionate to the strength of the LoRA. I typically kept the strength at 1.25 for the regular model and 1.00 for the HiDef model.

I used between 25 - 40 steps.

Prompting is key with this LoRA. If you ask for a six pack on the subject, it probably won't work well. You can indicate that you'd like the subject to be muscular, however I'd also avoid mentioning the subject "should be skinny."

Notes

• This LoRA is not meant to produce stockier or "dad bod" guys. If you want a stocky man, Flux.1 natively supports that just fine, and if you want a "dad bod", please check out this LoRA: https://civarchive.com/models/1339196/dadbod-flux

• Chest hair generation is typically a little better with this LoRA.

• This LoRA intentionally doesn't fix the poor quality male nipples. For more accurate nipples, see: https://civarchive.com/models/1160070/better-men-nipples-flux. For more puffy nipples, check out one of my other LoRAs in the Josh Collection: https://civarchive.com/models/1856087?modelVersionId=2491429. Both of these do alter the stomach area a little, so I'd keep them on the lower end of strength or maybe use an even lower combination of both.

• Unless I find major issues with this LoRA, I probably won't release another version of it.

• Training data was men between 20 - 40, so it works best for guys around those ages.

• Some minor bleeding from the sample data set occurs with the model. Use the regular model over the HiDef one to avoid this.