What Is It?

There are probably about a dozen LoRAs that aim to bring reality to Flux images. This is inspired by their work, and my own take on it. It is probably the last Flux model I will publish here. I wanted to see if I could use a minimal training style and a minimal but focused data set to train a model that could generate images not perfect - but close enough to real it could fool your grandparents. This is the result. Technically it's version 4 but the first and only one I will publish.

an amateur photo of a polish woman, 5'8, 110lbs, wearing a cute blouse, brown haired, glasses, sideswept bangs, sitting at a house party

a woman at a house party with a friend

all of the samples provided use only this LoRA and Face Detail node.

• euler/beta @ 32 steps OR

• res_3m/kl_optimal @ 28 steps.

• Model Strength: 0.80

Suggested Weight: 0.4 - 1.0

Training Notes

• Trained with 29 captioned images at 1024px resolution.

• No effort was made to train NSFW concepts, no nudity in the training set