FLUX 3 Image Klein

TL;DR

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Black Forest Labs has released FLUX 3 Image Klein, a 5.6-billion-parameter model for text-to-image generation, editing and multi-reference workflows. The weights are available under the FLUX Kommunity License, while BFL’s performance comparisons and safety findings are based on company-reported evaluations.

Black Forest Labs has released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model for generating and editing images, including workflows using as many as 10 reference images. The release gives developers downloadable base and distilled weights, and offers the model through the BFL API, making the company’s latest image tools available for local deployment as well as hosted use.

The company says Klein shares key capabilities with its larger FLUX 3 Image model, including native 4K image generation, pixel-level editing and bounding-box controls that can position elements or limit edits to a selected area. The multi-reference feature accepts up to 10 images, according to BFL. The release includes a base checkpoint intended for fine-tuning and a step-distilled version designed for lower-latency generation.

BFL also lists FP8, INT8 and NVFP4 builds, developed with NVIDIA support, alongside the standard BF16 weights. It says the BF16 version requires 11.2 GB of memory and the FP8 version 5.6 GB. With the INT8 build, the model can run on NVIDIA GeForce RTX GPUs with as little as 8 GB of VRAM, using ComfyUI or Diffusers, according to the company.

For speed, BFL reports that the distilled INT8 model can generate a one-megapixel image in about four seconds on an RTX 5090 and under nine seconds on an RTX 4070 through ComfyUI. These figures are for a 16-step run and were measured by NVIDIA, the report says. BFL also reports a sub-0.6-second result on NVIDIA GB300 hardware using NVFP4; that figure concerns a different GPU and build and should not be treated as a typical consumer-computer benchmark.

Local Image Tools on Smaller GPUs

Klein’s release matters to teams that want image generation and editing without sending every job to a hosted service. Open weights can be downloaded and run on local infrastructure, while the company’s smaller parameter count and quantized builds are intended to lower hardware demands. That may be relevant for organizations with data-control requirements and developers who want to adapt a model to a particular workflow.

The model also offers a choice between local use and a managed API. BFL says the API’s distilled model is available as flux-3-image-klein, with prompt upsampling and pixel restoration; the latter is described as preserving unedited image regions exactly. The company lists pricing from $0.009 per 1K image. The release could broaden access to editing and multi-reference generation, but the practical result will depend on image quality, speed on a user’s own hardware, and the license terms that apply to their organization.

Weights, Benchmarks and Licensing

FLUX 3 Image Klein is the second open-weights release in the FLUX 3 family, according to Black Forest Labs. The company says the model builds on shared training across images, video and audio, which it argues helps it represent spatial detail, motion, physical dynamics and events. Those descriptions are the developer’s account of the approach, not an independent assessment of how each training signal affects outputs.

BFL reports that Klein ranked first in its internal evaluations for text-to-image generation, multi-reference generation and editing, including comparisons with larger open models and the closed models Ideogram V4.5 and FLUX.2 [pro]. It also reports gains of 91 Elo points in text-to-image and 70 in editing over FLUX.2 [klein] 9B, despite using 38% fewer parameters. These are company-reported benchmark results; the supplied material does not provide enough independent testing detail to establish how the rankings translate across prompts and user tasks.

The weights are distributed under the FLUX Kommunity License. BFL says individuals and companies with annual revenue below $5 million may use them commercially for free, alongside research and non-commercial use. Larger organizations can use the API or pursue the company’s self-hosted licensing options. Support at release includes ComfyUI, Diffusers, TensorRT-LLM, SGLang Diffusion and vLLM-Omni, with LoRA training support through Ostris’s AI Toolkit and NVIDIA NeMo AutoModel recipes.

““Today we are releasing FLUX 3 Image Klein, a 5.6B parameter open weights model for text-to-image generation, image editing and multi-reference generation with up to 10 reference images.””

— Black Forest Labs

Independent Results Still Pending

The available performance figures and rankings come from BFL and its testing partners. The supplied material does not include independent benchmark results or enough methodological detail to determine whether its comparisons use equivalent prompts, settings and hardware across models. API latency comparisons also use different measurement methods: BFL says its own timing was measured on its servers, while competing models were timed from the client, including polling and network time.

Real-world performance will vary with GPU, quantization, software and workflow. It is also not yet clear from the announcement how consistently Klein handles the full range of editing requests, reference images or fine-tuned uses. BFL says it uses dataset filtering, targeted fine-tuning and red-team evaluations to address safety, and reports fewer vulnerabilities than other popular open-weights models; the supplied material does not identify those models or provide detailed evaluation results.

Testing on Local Hardware

The weights, reference inference code and a bounding-box demo are available through Hugging Face, while users can access the hosted version through the BFL API and Playground. Developers can test the base checkpoint for fine-tuning or the distilled checkpoint for faster generation, then compare results on their own hardware and workflows.

The next useful evidence will come from independent evaluations of image quality, editing accuracy, speed and safety across a wider set of prompts and devices. Users and organizations considering deployment will also need to check the license against their commercial use, especially if annual revenue exceeds the stated threshold. BFL has not provided a timetable for further FLUX 3 releases in the supplied announcement.

Key Questions

What is FLUX 3 Image Klein?

It is a 5.6-billion-parameter open-weights model from Black Forest Labs for text-to-image generation, image editing and multi-reference generation using up to 10 reference images.

Can it run on a consumer GPU?

BFL says its INT8 build can run on NVIDIA GeForce RTX GPUs with at least 8 GB of VRAM through ComfyUI or Diffusers. Speed and memory use may vary by hardware and setup.

Where can users get the model?

The base and distilled weights are listed on Hugging Face. The model is also available through the BFL API and Playground, with support for several inference and development tools.

Can companies use the weights commercially for free?

BFL says the FLUX Kommunity License permits free commercial use for individuals and companies with annual revenue below $5 million. Larger organizations can pursue self-hosted licensing or use the API; users should review the license for their specific use.

Are the performance claims independently verified?

The announcement describes internal evaluations and measurements by NVIDIA. The supplied source does not provide independent benchmark results, so the reported rankings and speeds should be understood as company-reported figures.

Source: hn