I Trained FLUX.1 on My Face (and how you can too)

Easy fine-tuning guide with Python and Replicate

Although large language models (LLMs) seem to get all the attention these days, image-generation models have been advancing just as rapidly. The current state-of-the-art is FLUX.1, an image model from Black Forest Labs (a faction from the Stable Diffusion team). In this article, I share the full process I used for fine-tuning this model to generate unlimited high-quality photos of myself.

Fine-tuned model outputs (with generous appropriations of hair and physique).

FLUX.1 is a multimodal AI model that can process text and images. More specifically, it takes text and image inputs and generates image outputs. This allows you to create images from text prompts alone or edit images using text!

The model comes in 3 versions: Pro, Dev, and Schnell (German for fast). Pro is the best-performing model, and Schnell is the quickest and cheapest.

Although these models can do incredible things, they can’t produce photos of specific people (unless they are well-known public figures). This is where fine-tuning can help us.