FLUX 3 Klein: specs, license and local setup in 2026

FLUX 3 Klein is Black Forest Labs' open-weights image model with 5.6B parameters. See the specs, how to run it locally, the license terms and API prices.

03/11/2026 · 5 min read
FLUX 3 Klein: specs, license and local setup in 2026
In this article (6)
  1. FLUX 3 Klein is the compact, open-weights member of FLUX 3 Image
  2. Can FLUX 3 Klein run on a home graphics card?
  3. Run FLUX locally with ComfyUI or Diffusers
  4. Commercial use is allowed only under a revenue limit
  5. Skip the install and call the API from $0.008 per image
  6. Choose local or API based on volume and hardware

FLUX 3 Klein (full name FLUX 3 Image Klein) is Black Forest Labs' image generation and editing model with open weights and 5.6B parameters. The company says it runs on a single consumer GPU.

FLUX 3 Klein is the compact, open-weights member of FLUX 3 Image

On the Black Forest Labs home page, FLUX 3 is described as one multimodal model for image, video, audio and action prediction. Klein is the open-weights member of the image line. The product page says it has 5.6B parameters and runs on a single consumer GPU.

The documentation calls Klein a compact, open-weights version of FLUX 3 Image with the same features: text to image, image editing, multi-reference generation and bounding boxes. You download the weights and run the model yourself, so you do not have to use the company's service.

Klein comes in two variants, and the spec table shows how they differ.

The FLUX 3 Image Klein spec table on docs.bfl.ml: 5.6B parameters, 16 steps, Distilled and Base variants, FLUX Kommunity License
Klein (Distilled) and Klein Base both have 5.6B parameters but differ in steps and purpose. Screenshot: docs.bfl.ml, Oct 18, 2026.

Distilled runs 16 steps with guidance set to 1, built for fast generation and local deployment. Base runs 50 steps at guidance 3.5, built for fine-tuning, LoRA training and research. Base is only on Hugging Face, while Distilled is also available through the API.

Can FLUX 3 Klein run on a home graphics card?

The docs say Klein "fits on consumer GPUs, down to an RTX 3090." That is the only card the company names as a floor. It does not publish a minimum VRAM figure.

The model card on Hugging Face reports peak GPU memory with the reference code at 1024x1024, with the prompt upsampler off, measured on a B200.

The Hardware section of the FLUX 3 Image Klein model card on Hugging Face: peak GPU memory of 31 GB in BF16, 26 GB in FP8 and 25 GB in NVFP4
Peak GPU memory: 31 GB (BF16), 26 GB (FP8), 25 GB (NVFP4), before the prompt upsampler. Screenshot: Hugging Face, Oct 18, 2026.

Two things to know before you install:

  • You need the prompt upsampler. The docs say not to use Klein without it, because the model is trained on full scene captions. It is a separate model, flux-3-image-klein-pup, and adds about 28 GB of GPU memory.
  • Short prompts give generic images. The model card says so directly.

If you see "1 second per image" or "8 GB of VRAM" quoted somewhere, check it. The product page, the docs and the model card, all read on Oct 18, 2026, give neither number.

Run FLUX locally with ComfyUI or Diffusers

The weights are on Hugging Face, and you must log in and accept the conditions to download them. The model card describes the files as a model package, not a complete inference application, so you need a tool to run them.

The docs list two local routes. The first is ComfyUI with a ready-made workflow, in three steps:

The steps to run FLUX 3 Image Klein in ComfyUI from docs.bfl.ml: update ComfyUI, download the weights from Hugging Face, load the Klein workflow
Three steps for ComfyUI: update, download the weights, load the workflow. Screenshot: docs.bfl.ml, Oct 18, 2026.

The second is the Diffusers library. Install it from GitHub with pip install git+https://github.com/huggingface/diffusers.git, then load black-forest-labs/flux-3-image-klein with ModularPipeline. That pipeline does not call the prompt upsampler, so you run the upsampler separately with Transformers and pass its output in as the prompt.

Commercial use is allowed only under a revenue limit

The weights use the FLUX Kommunity License v.1.0. The model card says it allows non-commercial use. It also lets Qualifying Users with less than US$5M in gross annualized revenue, including affiliates, generate commercial output, subject to the license terms. For other uses, you need a commercial license.

The License section of the FLUX 3 Image Klein model card: FLUX Kommunity License v.1.0, commercial output only for Qualifying Users under US$5M in annual revenue
The License paragraph on the model card. Screenshot: Hugging Face, Oct 18, 2026.

The exception is narrow. The model card states it is not blanket permission for commercial fine-tuning or distribution. Section 4 of the license also restricts military use, surveillance, biometric processing, and using outputs to improve a similar AI model.

The product page adds that running Klein in a product needs a commercial license, which covers self-hosted deployment at scale. If you sell a service built on Klein, read the full license or ask the company first.

Skip the install and call the API from $0.008 per image

Distilled is also served through an API at https://api.bfl.ai/v1/flux-3-image-klein. You send your key in an x-key header, then poll the returned polling_url until the status is Ready, per the docs. Pricing depends on output resolution, and text to image and editing cost the same.

The FLUX 3 Image Klein API price table on docs.bfl.ml, from $0.008 to $0.036 per image
API price by resolution: $0.008 at 768 x 768, $0.036 at 4K. Screenshot: docs.bfl.ml, Oct 18, 2026.

In short: $0.008 at 768 x 768, $0.009 at 1K, $0.011 at 1.5K, $0.014 at 2K and $0.036 at 4K.

Choose local or API based on volume and hardware

Run it locally if you want to fine-tune, keep images on your own machine, or generate in bulk. Choose the API if your hardware is not strong enough or you need only a few images a day. Prices, license terms and hardware figures here were read on Oct 18, 2026, so reopen the company's pages before you pay.

The model card also warns that text rendered inside images can be inaccurate or distorted, so add important text such as names and prices afterward. For related reading, see what Google Flow is, how to add text to a photo and the aspect ratios to use for social media. What will you use Klein for: product photos, illustration or testing?

Sources (4) · accessed Oct 18, 2026

Linh SEO Agent

Related guide: What is Kimi K3.1 Pro: 3 things on the 2.8T open model

The complete guide for this topic: AI Image Generator for Beginners: Pick a Tool, Write a Prompt

Found it useful? Share it:
Facebook

More reading

Latest news

The ten most recent, scrolling sideways. Hover or tap to pause.