FLUX: the new prodigy text-to-image model

In a context where generative artificial intelligence is growing exponentially, the text-to-image field never ceases to surprise. After the enthusiastic reception of AuraFlow, it's now Flux's turn to make its grand entrance on the scene of alternatives to Stability AI's Stable Diffusion.

Black Forest Labs, a new start-up specializing in generative AI for the media, has just launched Flux, its suite of text-to-image models featuring an open source approach coupled with impressive performance, making it a serious contender not only for Stable Diffusion 3 but also for other industry leaders such as DALL-E and Midjourney.

FLUX.1

Flux's text-image model, trained on a vast dataset of images and captions, has impressive capabilities for generating photorealistic images from natural language descriptions. The model's in-depth understanding of language enables it to interpret complex messages and produce highly detailed, coherent images.

Technical features

• Enhanced image quality: Create stunning images at higher resolutions.
• Advanced human anatomy and photorealism: Get highly realistic, anatomically accurate images.
• Faster adhesion: Get more accurate and relevant images based on your data.
• Exceptional speed: Benefit from the speed and efficiency of Flux Schnell, ideal for high-demand applications.
• Available in three variants:
  • FLUX.1 [dev]: The basic model, shared with a non-commercial license for the community to build on.
  • FLUX.1 [schnell]: A distilled version of the basic model, running up to 10 times faster under OpenSource license.
  • FLUX.1 [pro]: A private version available only via the API.

How to use FLUX.1

Online feeds

If you'd like to try out a few generations with FLUX.1, you can test the different versions on the Fal.AI and Replicate platforms:

Fal.AI

• FLUX.1 [pro]
• FLUX.1 [dev]
• FLUX.1 [schnell]

Replicate

• FLUX.1 [pro]
• FLUX.1 [dev]
• FLUX.1 [schnell]

The last two models are also shared on HuggingFace where they can be downloaded.

Download FLUX.1 [dev]

Download FLUX.1 [schnell]

The Black Forest Labs team has shared a sample code on Github that already allows developers and advanced users to run the models on their own machines.

Flux in ComfyUi

The latest update to ComfyUi should already integrate Flux and enable image generation with [dev] and [schell] models.

Workflows have been shared and are available on the Flux example page on the ComfyUi github.

Business strategy and licensing

Black Forest Labs takes an innovative strategic approach with its FLUX suite, offering options tailored to a variety of needs and usage contexts:

• FLUX.1 [pro] : The spearhead of the range, this high-end model promises unrivalled performance. Accessible only via the Black Forest Labs API, it targets professionals demanding exceptional visual quality. Although pricing details have yet to be confirmed, we can expect a premium commercial model.
• FLUX.1 [dev] : An intermediate version, FLUX.1 [dev] offers a balance between performance and accessibility. Designed for developers and researchers, it enables fine-tuning and more flexible use, while remaining framed by non-commercial terms of use. For commercial use, specific negotiations with Black Forest Labs will be necessary.
• FLUX.1 [schnell]: A true gateway to the FLUX universe, this speed-optimized version sacrifices some quality for accessibility. Distributed under an open source license (Apache 2.0), FLUX.1 [schnell] opens up a vast field of possibilities for large-scale experimentation and innovation.

This three-pronged strategy illustrates Black Forest Labs' determination to democratize access to generative AI while preserving a viable business model. It thus offers a range of solutions to meet the varied needs of professionals, researchers and AI enthusiasts alike.

Future prospects

The launch of FLUX by Black Forest Labs marks a turning point in the text-to-image industry. Backed by a $31 million Seed round led by Andreessen Horowitz, the company is well positioned to influence the future of generative AI.

The Black Forest Labs team, made up of renowned researchers and engineers who have contributed to major innovations such as VQGAN and Stable Diffusion, is already planning the development of text-to-video systems. This approach could accelerate innovation in fields as diverse as cinema, advertising and education.

By democratizing access to these cutting-edge technologies while emphasizing transparency and security, Black Forest Labs aims to shape a more open, collaborative and innovative generative AI ecosystem. Time will tell whether FLUX will be able to hold its own against the giants of the sector, but one thing is certain: competition in the field of text-to-image AI has only just begun.