FLUX 3 Klein 8GB VRAM: the open AI image model that runs on your PC
FLUX 3 Klein runs on 8GB of VRAM with open weights and no per-image fee. Here are the three numbers and the license fine print.
FLUX 3 Klein is an open-weights AI image model from Black Forest Labs, released on October 15, 2026. According to BFL, its INT8 version runs on RTX cards with as little as 8 GB of VRAM, so you can generate images on your own PC with no per-image fee.
What FLUX 3 Klein is
Black Forest Labs published the model with open weights, in a base and a distilled version, on Hugging Face. It is also available through the BFL API, and from day one it works with ComfyUI, Diffusers, TensorRT-LLM, SGLang and vLLM-Omni.
The three numbers
- Memory: the INT8 version needs as little as 8 GB of VRAM on an NVIDIA GeForce RTX card, through ComfyUI or Diffusers.
- Speed: about 4 seconds for a 1 MP image on an RTX 5090, and under 9 seconds on an RTX 4070 in ComfyUI.
- Size: 5.6 billion parameters, 38% fewer than FLUX.2 [klein] 9B.
The speed figures and the size comparison are BFL's own numbers, not an independent test. The 8 GB figure applies only to the INT8 version, and the 4 seconds is on a top-tier card.
The license fine print
Commercial use is free for individuals and for companies with annual revenue under $5 million, under the FLUX Kommunity license. Research and non-commercial use are covered too. Larger organizations need the self-hosting license or the API.
How to apply it in your business?
If you generate product visuals, social posts or mockups, running the model on a PC you already own replaces a per-image charge with a one-time hardware cost. Check your company's annual revenue against the $5 million threshold first, because above it the license changes. Then test the INT8 version in ComfyUI with your most common image type before committing.
Want to see another open model that runs on local hardware? Read the full breakdown of Muse Spark Open on two Mac Studios. For image pricing on the paid side, see the Nano Banana 2.1 pricing breakdown.
FAQ
Can FLUX 3 Klein really run on 8 GB of VRAM?
Yes, according to BFL, but only the INT8 version, on NVIDIA GeForce RTX cards through ComfyUI or Diffusers.
Is FLUX 3 Klein free for commercial use?
It is free for individuals and for companies with annual revenue under $5 million. Above that, you need another license.
How fast is FLUX 3 Klein?
BFL reports about 4 seconds for a 1 MP image on an RTX 5090 and under 9 seconds on an RTX 4070.
Does it have open weights?
Yes. The base and distilled versions are published on Hugging Face.
Conclusion
FLUX 3 Klein with 8 GB of VRAM puts an open image model on ordinary gaming hardware, without a meter running. The catch is the fine print: free commercial use stops at $5 million in annual revenue, and the speed claims come from the maker. Tell us what image you generate most and whether your PC can run it.