Black Forest Labs releases FLUX1.1 [pro] and its API

On 2 October 2024, Black Forest Labs released FLUX1.1 [pro] and made its BFL API generally available.^[1] The two announcements came together: the model was the product, and the API was the way most developers would reach it.

The model

A text-to-image model takes a written prompt and returns an image that matches it. FLUX1.1 [pro] was a new version in Black Forest Labs' FLUX line of such models, and the announcement centred on how much faster it ran and how developers could pay for it.

Speed, and a moving baseline

Black Forest Labs said the new model was six times faster than FLUX.1 [pro].^[1] A month later the company gave a smaller number: FLUX1.1 [pro] was three times faster than the FLUX.1 [pro] then available, because FLUX.1 [pro] had itself been updated to run twice as fast.^[3] The two figures describe different comparisons rather than a contradiction — the sixfold claim was measured against the older release, the threefold claim against the updated one. Speed comparisons of this kind depend entirely on which earlier version is used as the baseline, and when a model line is itself getting faster, both statements can be true at once.

Ranking under a codename

At release, a model listed under the codename "blueberry" led the Artificial Analysis image arena.^[1] Arenas of this kind typically rank models by showing users two images made from the same prompt and asking which they prefer, without saying which model produced which. A model's position therefore reflects pairwise preferences rather than a fixed benchmark score. Entering an arena under a codename, before the model is named publicly, is a common way to test it.

Price and access

API pricing was 4 cents per image.^[1] An application programming interface (API) lets software send a prompt to a hosted model and receive the generated image back. Per-image pricing turns image generation into a metered cost that a developer can budget for, without running the model on its own hardware. Black Forest Labs also said FLUX1.1 [pro] would be available through Together.ai, Replicate, fal.ai and Freepik,^[3] so developers had third-party routes to the same model alongside the official API.

Later modes

On 6 November 2024, Ultra and Raw modes followed.^[2] Offering one model in several modes is a way to let users pick an output setting for the job at hand.

Why it matters

The release put the pieces of a text-to-image model launch together in one place: a named model version, a speed claim, a leaderboard position, a per-image price and a set of hosting partners. For a developer deciding whether to build on a model, those pieces answer different questions — how fast it runs, how good it looks to users, what each call costs, and where it can be reached from. Making the API generally available also changes what the model is for: it becomes a component that other software can call, which is a different role from a demo in a browser.