Black Forest Labs launches FLUX.2 and Kontext via API, withholds open weights

Black Forest Labs has released FLUX.2 and FLUX Kontext through its public API, marking a strategic departure from the open-weight distribution model that defined FLUX.1.

FLUX.2 is positioned as the successor to the original FLUX weights, claiming state-of-the-art quality, speed, and controllability for AI image synthesis. The Kontext variant suggests context-aware or conditional generation capabilities. Both models are now available through the company's API catalog at bfl.ai. Black Forest Labs has not yet published parameter counts, context-window sizes, benchmark comparisons, technical specifications, pricing tiers, per-image costs, or throughput limits.

The original FLUX.1 [dev] and FLUX.1 [schnell] weights, released in mid-2024, quickly became fixtures in the local image generation ecosystem. The schnell variant optimized for speed at four-step sampling, while the dev checkpoint prioritized output fidelity. Both topped community leaderboards for prompt adherence and photorealism, and their permissive licenses enabled widespread fine-tuning in ComfyUI and A1111 workflows. FLUX.1 became a go-to base for LoRA training, particularly for photorealism and character-consistency tasks.

The API-only release of FLUX.2 leaves a critical gap for practitioners. Those accustomed to running FLUX locally on consumer RTX 4090 or professional A6000 cards now face a choice: pay per-image API costs or wait for downloadable checkpoints that may never arrive. The absence of open weights also complicates fine-tuning workflows that rely on direct parameter access—a cornerstone of FLUX.1's community uptake. Black Forest Labs has not published a changelog, release notes, or technical paper accompanying the launch, leaving performance claims unverified against competitors like Midjourney, DALL·E 3, or Stable Diffusion 3.5.