Black Forest Labs (BFL) today released the open weights for FLUX 3 Image Klein, its most capable downloadable image model so far — a 5.6-billion-parameter model that runs on consumer GPUs with as little as 8 GB of VRAM. Why it matters: frontier-quality image generation and editing no longer require an API account or a data center.
October 15, 2026 — Freiburg, Germany. Black Forest Labs, the lab founded by the researchers behind Stable Diffusion, announced the open weights release of FLUX 3 Image Klein on Thursday. The model is the creative-focused member of the FLUX 3 family, following September’s open weights release of FLUX 3 Action, which targets robotics.
What was released
The release includes a base model intended for fine-tuning and research, plus a step-distilled version that generates images in fewer sampling steps for lower latency. Quantized FP8, NVFP4, and INT8 builds are also available, which is how the model fits comfortably on laptop-class hardware. Klein keeps several features from the larger, API-only FLUX 3 Image model, including native 4K generation, bounding-box control, and pixel-level editing, and it supports text-to-image generation, image editing, and generation from multiple reference images.
In BFL’s own internal evaluations — company-run numbers, not independently verified — Klein scored ahead of other current open weights image models on text-to-image, multi-reference generation, and image editing, while costing less through BFL’s API than comparable options. Treat those rankings as the company’s claim until third-party benchmarks land.
The license, in plain terms
Klein ships under the FLUX Kommunity License: free commercial use for individuals and companies with annual revenue under $5 million, plus research and non-commercial use. Larger organizations need BFL’s self-hosted licensing or the paid API. Launch-day tooling support includes Ostris’s AI Toolkit for fine-tuning, ComfyUI, Hugging Face Diffusers, vLLM-Omni, and SGLang Diffusion, and the model was trained on NVIDIA GB200 NVL72 systems and optimized for GeForce RTX cards.
Why it matters
Open weights change who controls the workflow. Teams with data-sovereignty rules, privacy-sensitive clients, or per-image API bills they want to kill can now run a frontier-grade image model entirely on their own hardware, fine-tune it on their own examples, and keep every prompt and output in-house. BFL says its models have passed 640 million downloads on Hugging Face, and Klein looks built to extend that lead: the practical takeaway for creators and small studios is that a genuinely capable local image model just got smaller, cheaper to run, and legally usable in commercial work under the revenue cap. As always with open weights, the safeguard story matters too — BFL says external partner Cinder evaluated both checkpoints before release and found fewer vulnerabilities than other popular open models.
Frequently asked questions
Can FLUX 3 Image Klein run on my PC? If you have a GPU with at least 8 GB of VRAM, yes — the quantized builds are designed for consumer GPUs and laptops.
Can I use it commercially for free? Yes, if your annual revenue is under $5 million, under the FLUX Kommunity License. Larger companies need a self-hosted license or the API.
How is Klein different from FLUX 3 Image? Klein is the smaller, open-weights sibling. It retains headline features like native 4K and pixel-level editing but is sized for local hardware rather than maximum quality.
Sources: Black Forest Labs announcement via GlobeNewswire (October 15, 2026).