FLUX.1

What it is

FLUX.1 is a text-to-image model family from Black Forest Labs. It turns a written description into an image ^[1]. The family was released on 1 August 2024 alongside the company's launch and its $31 million seed round ^[1].

Origins

Black Forest Labs released the FLUX.1 text-to-image family on 1 August 2024 ^[1]. The company was founded by Robin Rombach, Patrick Esser and Andreas Blattmann, who are described as the original creators of Stable Diffusion ^[2]. The company's later model, FLUX1.1 [pro], followed on 2 October 2024 and is the successor to FLUX.1 ^[4].

Key features

FLUX.1 comes in three variants. FLUX.1 [pro] is available through the API only. FLUX.1 [dev] is open-weight and guidance-distilled, meaning it is trained to produce results that would otherwise need extra guidance steps, and it carries a non-commercial license. FLUX.1 [schnell] is the fastest variant and is released under the Apache 2.0 license ^[1].

All three variants have 12 billion parameters. A parameter is one of the numbers a model learns during training, so 12 billion parameters refers to the size of the network ^[2].

The architecture is a hybrid of multimodal and parallel diffusion transformer blocks, trained with flow matching. The design uses rotary positional embeddings, which encode the position of each element in the input, and parallel attention layers ^[1]. Flow matching is a training method that learns a path from noise to images.

The models support aspect ratios and resolutions from 0.1 to 2.0 megapixels ^[1]. A megapixel is one million pixels, so this range sets the smallest and largest image sizes the models handle.

At launch, [pro] and [dev] were available on Replicate and fal.ai, with [dev] weights on Hugging Face. [schnell] was released on GitHub and Hugging Face Diffusers, with day-one support in ComfyUI ^[1].

Black Forest Labs said [pro] and [dev] surpassed Midjourney v6.0, DALL·E 3 (HD) and SD3-Ultra in visual quality, prompt following, size and aspect variability, typography and output diversity ^[1]. These are the company's own claims at launch, not independent test results.

Developments after launch

In August 2024 FLUX was integrated into Grok on X as a premium image feature ^[3]. On 21 November 2024 the company announced FLUX.1 Tools, which include Fill, Depth, Canny and Redux, for editing and control ^[3].

In October 2024 the company announced API prices: 2.5 cents per image for FLUX.1 [dev] and 5 cents for FLUX.1 [pro] ^[4]. The successor, FLUX1.1 [pro], was announced on 2 October 2024 ^[4].

Place in development

FLUX.1 was the model family released alongside the company's launch ^[1]. It set the tiered pattern that the company kept after launch: a hosted [pro] model, an open-weight [dev] model and an open-weight [schnell] model ^[1][4].