The Bottom Line: The company behind the FLUX family of visual intelligence models. Its latest FLUX 3 is a single multimodal model spanning image, video, and audio generation—with up to 20-second video clips, synchronized audio, and keyframe control from text or image inputs. The FLUX.2 image family (klein/pro/flex/max) offers a spectrum from sub-second inference to highest-quality production assets, with multi-reference editing up to 10 images and grounding search on max. BFL offers three access paths: API (credit-based, 1 credit = $0.01), open weights (klein 4B under Apache 2.0, 9B under non-commercial license), and enterprise licensing. An official MCP server lets Claude, Cursor, and other MCP clients generate and edit FLUX images directly. FLUX.2 [pro] received a ~2x speed upgrade in March 2026 with no price change.

Pricing model: Freemium — open weights for klein 4B are free for commercial use under Apache 2.0 . API access is credit-based at 1 credit = $0.01, with FLUX 3 Image ranging from $0.041 (768×768) to $0.607 (4K), and FLUX 3 Video billed per second depending on resolution and mode . Open-weight licensing tiers start at Builder (10K images/month) and scale to Enterprise . The 9B klein model requires a commercial license for business use .

Platforms: Web (API platform, Playground, dashboard). No native desktop or mobile app for end users is documented. Access is through API, MCP server, and third-party platforms like fal.ai .

Pros & Cons

Pros: The FLUX family spans an unusually wide capability range—from klein 4B (sub-second, runs on ~13GB VRAM, Apache 2.0 open weights) to max (highest quality, grounding search, multi-reference up to 10 images), covering real-time experimentation through production-grade output . The official MCP server brings FLUX generation and editing into Claude, Cursor, Codex, and Windsurf with OAuth sign-in and no API key pasting . FLUX 3 consolidates image, video, and audio into one model with native synchronized audio and up to 20-second clips. Pricing is transparent and granular, with a published rate card covering every model and resolution tier . LoRA inference is self-serve on FLUX.2 klein via managed fine-tuned endpoints.

Cons: The most capable models are not open—klein 4B is Apache 2.0, but klein 9B is non-commercial only, and pro/flex/max are API-only, so the "open weights" positioning applies only to the entry tier . Video generation (FLUX 3) is relatively new, and independent large-scale validation of quality versus dedicated video models is thin; Artificial Analysis places FLUX 3 Max at 1,138 Elo on image-to-video, essentially tied with Seedance 2.5 within overlapping confidence intervals . Credit-based pricing requires monitoring for high-volume generation, especially at 4K image and UHD video rates . The company's identity has shifted from "the open Stable Diffusion successor" to a multimodal commercial platform. No native mobile app or consumer-facing product—BFL is developer and enterprise infrastructure.

Who is it for?

Ideal for: Developers and teams that want production-grade image generation with API access and transparent per-image pricing; open-weight enthusiasts who need a capable image model (klein 4B, Apache 2.0) for local deployment, fine-tuning, or LoRA training; teams exploring multimodal generation (image + short video + audio) from a single vendor; Claude/Cursor users who want FLUX capabilities via MCP without building an integration; enterprises requiring enterprise licensing for pro/flex/max models or custom deployment.