Black Forest Labs Releases FLUX 3 Image Klein: 5.6B Open-Weights Image Model Runs on 8GB GPUs

What happened

On Oct 15, Black Forest Labs released FLUX 3 Image Klein with open weights. The model supports text-to-image generation, pixel-level multi-turn editing of existing images, and composition using up to 10 reference images in a single generation. The company calls it its most capable open-weights image model to date.

Klein ships in three forms: a base checkpoint for fine-tuning and research, a distilled version with fewer steps (16 steps, low latency), and FP8, NVFP4 and INT8 quantized builds developed with NVIDIA. Local workflows support ComfyUI and Hugging Face Diffusers, with LoRA training available from day one through Ostris AI Toolkit and NVIDIA NeMo AutoModel.

FLUX 3 allows that community to adapt the same foundation to very different needs, from creative generation to robotics.

— Robin Rombach, co-founder and CEO of Black Forest Labs (company announcement, reported by VentureBeat)

Key facts

1. 5.6B parameters: 38% fewer than the 9B FLUX.2 [klein]; the company's internal evaluation claims a 91-point higher text-to-image Elo (reported as company-tested figures)
2. Runs on 8GB VRAM: the INT8 build runs on GeForce RTX cards with as little as 8GB VRAM; the company reports about 8.9 seconds per 1024×1024 image on an RTX 4070
3. License: FLUX Kommunity License — free commercial use for individuals and businesses with under $5M annual revenue; larger organizations need a separate self-hosting arrangement or the API
4. API pricing: $0.009 per 1K image on the BFL API, the same for generation and editing; $0.036 at 4K
5. Safety review: the company says external partner Cinder evaluated both checkpoints before release