FLUX 3 Image Klein: Open 4K Image AI That Runs on 8GB GPUs

FLUX 3 Image Klein brings high-end AI image generation to ordinary gaming PCs. Released on October 15 by Black Forest Labs, the 5.6-billion-parameter open-weights model handles text-to-image generation, precise editing and multi-reference composition, and its INT8 build runs on NVIDIA GeForce RTX graphics cards with as little as 8GB of VRAM. For hobbyists and small studios, that puts native 4K output on hardware many people already own.

• Model size: 5.6B parameters, with weights of 11.2GB in BF16 and 5.6GB in FP8.
• Capabilities: native 4K generation, pixel-precise editing, bounding-box layout control and up to 10 reference images.
• Speed: about 4 seconds per 1-megapixel image on an RTX 5090 and under 9 seconds on an RTX 4070 with the distilled INT8 build.
• License: free commercial use for individuals and companies earning under $5 million a year.

What Is FLUX 3 Image Klein?

Klein is the compact member of the FLUX 3 family and the second open-weights release in it, following FLUX 3 Action. It shares the flagship's core features, including bounding boxes that let you pin an element to a specific spot or confine an edit to one region of a picture. If you followed our coverage of FLUX 3 Video reaching general availability, Klein is the image-side counterpart sized for local hardware.

Black Forest Labs ships two variants. The base model, run at 50 steps in the company's benchmarks, is meant for fine-tuning and LoRA training. The step-distilled model runs at 16 steps for fast generation. Both come in FP8, INT8 and NVFP4 builds developed with NVIDIA.

Can FLUX 3 Image Klein Run on My GPU?

Very likely, if you have a recent GeForce card. The INT8 build is the one to grab for 8GB cards through ComfyUI or Hugging Face Diffusers. Black Forest Labs' published timings for a 1024 x 1024 image look like this:

• RTX 4070 (distilled, INT8): under 9 seconds.
• RTX 5090 (distilled, INT8): about 4 seconds.
• GB300 data-center GPU (distilled, NVFP4): under 0.6 seconds.
• DGX Spark (base model, 50 steps): 43.8 seconds.

Adding a LoRA increases generation time by roughly 5%, according to the company, so custom styles stay practical for local AI workflows.

How Does Klein Compare to Larger Models?

In Black Forest Labs' internal Elo evaluations, Klein gains 91 points in text-to-image and 70 points in editing over the earlier FLUX.2 [klein] 9B model, despite having 38% fewer parameters. The company also says it ranks ahead of several larger open models and some closed services in its tests. These are vendor results, so expect community comparisons in the coming weeks, but the size-to-quality trend is clear: image models keep getting smaller without getting worse.

Pricing, License and Day-One Tools

The FLUX Kommunity License makes Klein free for commercial use by individuals and businesses under $5 million in annual revenue, as well as for research and non-commercial work. Larger companies can license self-hosting or use the API, which VentureBeat reports is priced at about $0.009 per image at 1K resolution and $0.036 at 4K.

Day-one support covers ComfyUI, Diffusers, TensorRT-LLM, SGLang Diffusion and vLLM-Omni, and LoRA training works through Ostris's AI Toolkit and NVIDIA NeMo recipes. Black Forest Labs says the release was red-teamed internally and evaluated by external partner Cinder before launch.

A capable 4K image model that fits on an 8GB gaming card is a genuine step forward for open creative tools. Find more releases like this in our AI news section.

Sources: Black Forest Labs, FLUX 3 Image Klein — October 15, 2026; VentureBeat, Black Forest Labs opens FLUX 3 Image Klein weights — October 15, 2026; Hugging Face, FLUX 3 Image Klein — October 15, 2026.