What can FLUX do?

The team that developed Stable Diffusion left Stability AI to form a new company called Black Forest Labs.

Black Forest Labs released a family of open-access image generation AI models called FLUX.
The most powerful model in the family, available for on-premises installations, is called FLUX 1 Dev.

Inspecting the pictures shared on social media like Reddit and X, it’s immediately obvious that FLUX 1 Dev is more capable than Stable Diffusion models (SD 1.5, SDXL, and the unfortunate SD3 Medium) and their fine-tuned variants.

What is not immediately obvious is how far FLUX 1 Dev can go, especially in comparison to the gold standard in image generation: Midjourney.

After almost two years of exposure to Stable Diffusion images, creative studios and artists don’t expect open-access models to be competitive.

Hence, the risk is that the creative industry won’t investigate the full potential of newly released models.

This post tries to document the full potential of FLUX 1 Dev, with a focus on commercial and creative imagery. As such, it will be constantly updated, as the AI community discovers new capabilities.

The content is divided into two categories:

Unless otherwise specified, all images you’ll see in this post have been generated with AP Workflow for ComfyUI.

Control

The capability to control what an AI model does (often called prompt adherence) is critical in the artistic creation process.

Stable Diffusion models are notoriously hard to control. Both version 1.5 and XL require users to learn a complicated, sometimes unreasonable, sequence of incantations.

FLUX 1 Dev is much easier to instruct and much better at following instructions (the prompt).

To test the prompt adherence capabilities of various image-generation AI models, Alessandro created a series of challenges.

Below you’ll find the list of challenges that FLUX 1 Dev passed.

The Seventh Seal Challenge

…

The Medusa Challenge

…

Creativity

Another thing Stable Diffusion models are (in)famous for, especially compared to Midjourney, is the lack of creativity.

The images generated with SD 1.5 and XL can be aesthetically pleasing, but it’s extremely hard to achieve creative composition and style. In some cases, it’s outright impossible.

FLUX 1 Dev is a big step forward in this area.

Below you’ll see a series of examples, curated by Alessandro, mean to showcase its creative potential.

…