FLUX

Overview

FLUX is a family of text-to-image and image-editing models from Black Forest Labs (BFL), a Freiburg, Germany company founded in 2024 by Robin Rombach, Andreas Blattmann, and Patrick Esser, three former Stability AI researchers who worked on Stable Diffusion. The models are rectified-flow transformers (the original FLUX.1 line scaled to 12 billion parameters). The line launched with FLUX.1 in August 2024 and was extended by FLUX.2 on November 25, 2025. BFL has raised over $450M in total, including a $300M Series B in December 2025 at a reported $3.25B valuation co-led by Salesforce Ventures and Anjney Midha, with a16z, NVIDIA, General Catalyst, and Temasek participating.

What it does

FLUX generates images from natural-language prompts and edits existing images. The original FLUX.1 line ships as FLUX.1 [schnell] (fast, Apache 2.0 open weights), FLUX.1 [dev] (open weights, non-commercial), and API-only FLUX 1.1 Pro and 1.1 Pro Ultra (up to 4MP). The open toolset adds Fill (in/out-painting), Canny and Depth (structural conditioning), and Redux (image variation). FLUX.1 Kontext (June 2025) does multi-turn, in-context editing that keeps characters, identities, and styles consistent across edits, shipping as API-only Pro and Max plus an open-weight Kontext [dev]. FLUX.2 (November 2025) adds state-of-the-art in-context generation and editing with multi-reference support, shipping as FLUX.2 [klein] (Apache 2.0), FLUX.2 [dev] (non-commercial open weights), and proprietary Flex/Pro/Max API tiers; FLUX.2 reportedly uses latent flow matching with a Mistral-3 vision-language model. Throughout, a person prompts, curates, and refines: FLUX is an engine and assistant, not an autonomous agent.

Integrations & setup

The proprietary tiers are served through the BFL REST API (docs.bfl.ml). The open-weight variants can be downloaded from Hugging Face and run self-hosted via the Diffusers library and ComfyUI, and are served by third parties including Replicate, fal, Together AI, and the Vercel AI Gateway, so the models can be embedded in other products. The official inference repo (github.com/black-forest-labs/flux) provides minimal code to run the open models. FLUX exposes a REST API; no MCP or A2A support is documented as of this review.

Pricing

Usage-based, billed per image, with no subscriptions or seat fees (1 credit = $0.01). Reported per-image prices include FLUX 1.1 Pro at $0.04, 1.1 Pro Ultra at $0.06, FLUX.1 Kontext Pro at $0.04 and Kontext Max at $0.08, Fill Pro at $0.05, and FLUX.2 tiers from about $0.014 (Klein 4B) and $0.03 (Pro) up to $0.07 (Max), with FLUX.2 priced per megapixel. The open weights are free to download and run; [schnell] and [klein] are Apache 2.0, while [dev] variants are non-commercial and require a paid BFL license for commercial self-hosting. There is no documented free API tier. Always confirm current numbers on the pricing page.

Best for / not for

Best for developers and studios that want strong open-weight image models they can self-host (FLUX.1 [schnell], FLUX.2 [klein]), a pay-as-you-go API for production, and in-context editing (Kontext, FLUX.2) with consistent characters and styles. Less suited to non-technical users wanting a polished one-click consumer app (third-party front-ends are the easier path), to anyone wanting an autonomous agent, or to commercial teams that need open weights for the top quality tiers (those are proprietary and API-only).

Alternatives

Stable Diffusion (from the founders' former employer, Stability AI) is the other major open-weight image family with a comparable self-host ecosystem; Midjourney offers a more opinionated, high-aesthetic hosted app but no open weights; Ideogram and Recraft target in-image text and brand/vector design respectively. FLUX differentiates on combining open-weight tiers with strong in-context editing and a per-image API.