FLUX 3 Image Klein — what it is, how to run it, price & license

The launch video · Black Forest Labs on X, October 15, 2026

FLUX 3 Image Klein is here.

Open Weights with the quality of a large frontier model. Customizable and fast.

5.6B parameters and runs on a single consumer GPU.

Native 4K, bounding boxes, and pixel-perfect editing.
— Black Forest Labs (@bfl_ai) October 15, 2026

5.6Bparameters

4Knative output

10reference images

8 GBmin. VRAM (INT8)

$0.009per 1K image (API)

Capabilities

One small model. Generate, edit and compose.

Text-to-image

Photoreal, illustration, anime, typography — with legible text inside the image, up to native 4K.

Pixel-perfect editing

Change one thing with an instruction. Through the API, untouched pixels stay exactly the same.

Up to 10 references

Combine people, products, styles and places from up to 10 images into one new picture.

Bounding boxes

Say exactly where things go on a 0–1000 grid: place, move, recolor or remove elements. Format →

Runs on your GPU

INT8 build on 8 GB GeForce RTX cards; ≈4 s per image on an RTX 5090. Setup →

Free commercial use

FLUX Kommunity License: free for individuals and companies under $5M revenue. License →

At a glance

FLUX 3 Image Klein specs

Everything Black Forest Labs has published so far, in one table. Benchmarks are BFL's own.

Read the full explainer →

Made by	Black Forest Labs (BFL), Freiburg, Germany and San Francisco
Released	October 15, 2026 (open weights)
Size	5.6B-parameter image generator; text encoder Qwen3-VL-4B-Instruct (separate)
Does	Text-to-image, image editing, multi-reference generation (up to 10 reference images)
Output	Up to native 4K; text rendering in images
Control	Bounding boxes on a 0–1000 grid to place, move, recolor, replace or remove elements
Checkpoints	Klein (step-distilled, 16 steps, no guidance) and Klein Base (50 steps, guidance 3.5, for fine-tuning); FP8, NVFP4 and INT8 builds
Weights size	11.2 GB in BF16, 5.6 GB in FP8 (generator only)
Minimum GPU	NVIDIA GeForce RTX with 8 GB VRAM (INT8 build, ComfyUI or Diffusers)
Speed	≈4 s per 1 MP image on RTX 5090, under 9 s on RTX 4070 (distilled + INT8, BFL/NVIDIA figures)
API	Model id flux-3-image-klein, from $0.009 per 1K image
License	FLUX Kommunity License — free commercial use under $5M annual revenue
Runs in	ComfyUI, Hugging Face Diffusers, TensorRT-LLM, SGLang Diffusion, vLLM-Omni; LoRA training in Ostris AI Toolkit and NVIDIA NeMo AutoModel

Price

Five times cheaper than FLUX 3 Image

FLUX 3 Image Klein$0.009per 1K image · $0.036 at 4K

FLUX 3 Image$0.048per 1K image · $0.607 at 4K

Nano Banana 2.1$0.034per 1K image · $0.113 at 4K

FLUX.2 [klein] 9B$0.015per image

API list prices on October 15, 2026. All resolutions and sources →

What's new

vs FLUX.2 [klein]: +91 Elo, 38% fewer parameters

FLUX 3 Image Klein keeps the big FLUX 3 Image model's native 4K output, bounding boxes and pixel-precise editing, and adds up to 10 reference images. In BFL's own blind evaluations it scores +91 Elo in text-to-image and +70 Elo in image editing over FLUX.2 [klein] 9B. Full comparison →

Guides

Start using FLUX 3 Image Klein

Try the free demoOfficial Hugging Face Space — text-to-image, edits, bounding boxes.huggingface.co ↗ Run it locallyComfyUI or Diffusers, copy-paste code, VRAM and speed tables.Guide → API pricingModel id flux-3-image-klein, $0.008–$0.036 per image.Guide → LicenseThe $5M rule, what's allowed and what isn't.Guide → Prompting guideLong prompts, references, text in images, edits.Guide → Bounding boxesThe JSON format with ready-to-copy examples.Guide →

FAQ

Questions about FLUX 3 Image Klein

What is FLUX 3 Image Klein?

FLUX 3 Image Klein is a 5.6-billion-parameter open-weights image model from Black Forest Labs, released on October 15, 2026. One model does text-to-image, image editing and multi-reference generation with up to 10 reference images, at up to native 4K, with bounding boxes to control where things go — and its INT8 build runs on NVIDIA GeForce RTX GPUs with as little as 8 GB of VRAM.

Is FLUX 3 Image Klein free?

The weights are free to download from Hugging Face (after accepting the license) and free to use for research, non-commercial work, and commercial work by individuals and companies with under $5 million in annual revenue (FLUX Kommunity License). Larger companies need a self-hosting license from Black Forest Labs or can use the paid API. There is also a free demo on Hugging Face Spaces.

How much VRAM does FLUX 3 Image Klein need?

Black Forest Labs says the INT8 build runs on NVIDIA GeForce RTX GPUs with as little as 8 GB of VRAM through ComfyUI or Diffusers. The full BF16 reference code is much heavier: the model card measures about 31 GB peak (25–26 GB for NVFP4/FP8) on a data-center GPU at 1024×1024.

How fast is FLUX 3 Image Klein?

With the distilled model and the INT8 build in ComfyUI, a 1-megapixel image takes about 4 seconds on an RTX 5090 and under 9 seconds on an RTX 4070 (NVIDIA measurements published by BFL). Through the BFL API, median end-to-end latency at 1 MP is 4.54 s for text-to-image and 7.18 s for editing.

How much does the FLUX 3 Image Klein API cost?

Per image, the same for generation and editing: $0.008 at 768×768, $0.009 at 1K, $0.011 at 1.5K, $0.014 at 2K and $0.036 at 4K (as reported by VentureBeat from BFL's pricing).

Can I use FLUX 3 Image Klein commercially?

Yes if you are an individual or a company with less than US$5 million in gross annualized revenue (including affiliates), under the FLUX Kommunity License. Above that, use the BFL API or buy a self-hosting license.

Where can I download FLUX 3 Image Klein?

From Hugging Face: black-forest-labs/flux-3-image-klein (distilled, includes FP8/NVFP4 builds, the text encoder and autoencoder) and black-forest-labs/flux-3-image-klein-base (for fine-tuning). Reference code is on GitHub at black-forest-labs/flux-3-image-klein; a ComfyUI-packaged repo is black-forest-labs/flux-3-image-klein-comfy.

Make images with FLUX 3 Image Klein — no setup

ImageKlein puts the model in your browser: text-to-image, edits and up to 10 references at up to 4K.

Start generating

Sources