Black Forest Labs launches open source Flux.2 [klein] to generate AI images in less than a second

German AI startup Black Forest Labs (BFL), founded by former Stability AI engineers, continues to build out its suite of open source AI image generators with the release of FLUX.2 [klein]a new pair of small models – one open and one non-commercial – that emphasize speed and lower computing requirements, with the models generating images in less than a second on an Nvidia GB200.

The [klein] The series, released yesterday, includes two primary parameter counts: 4 billion (4B) and 9 billion (9B).

The model weights are available at Hugging face and code on Github.

While the larger models in the FLUX.2 family ([max] And [pro]), released November 2025, pushes the boundaries of photorealism and grounding search capabilities, [klein] is specifically designed for consumer hardware and latency-critical workflows.

Good news for businesses is that the 4B version is available under an Apache 2.0 license, which means they – or any organization or developer – can use the [klein] models for their commercial purposes without paying BFL or any intermediaries a cent.

However, there are a number of AI image and media creation platforms including Fal.ai have also started offering it at extremely low costs through their application programming interfaces (APIs) and as a direct-to-user tool. It has already received high praise from early users for its speed. What it lacks in overall image quality, it appears to make up for in its fast generation capabilities, open licensing, affordability, and small footprint – benefiting companies looking to run image models on their own hardware or at extremely low cost.

So how did BFL do it and how can it benefit you? Read on for more information.

The ‘Pareto frontier’ of latency

The technical philosophy behind it [klein] is what the BFL documentation describes as defining the “Pareto frontier” for quality versus latency. In simple terms, they’ve tried to squeeze the maximum possible visual fidelity into a model small enough to run on a home gaming PC without noticeable lag.

The performance metrics released by the company paint a picture of a model built for interactivity rather than just batch generation.

According to the official figures from Black Forest Labs, the [klein] models are capable of generating or editing images in less than 0.5 seconds on modern hardware.

Even on standard consumer GPUs like an RTX 3090 or 4070, the 4B model is designed to fit comfortably within about 13 GB of VRAM.

This speed is achieved through ‘distillation’, a process in which a larger, more complex model ‘teaches’ a smaller, more efficient model to approximate its results in fewer steps. The spirit [klein] variants require only four steps to generate an image. This effectively changes the generation process from a coffee break task to a near-instant task, enabling what BFL on

Under the hood: uniform architecture

Historically, image generation and editing often required various pipelines or complex adapters (such as ControlNets). FLUX.2 [klein] tries to unite them.

The architecture natively supports text-to-image, single-reference editing, and multi-reference composition without the need to switch models.

According to the documentation released on GitHub, the models support:

• Edit with multiple references: Users can upload up to four reference images (or ten on the playground) to determine the style or structure of the output.

• Color control with hexadecimal code: A common pain point for designers is achieving ‘just that shade of red’. The new models accept specific hexadecimal codes in prompts (e.g. #800020) to force accurate color reproduction.

• Structured prompting: The model parses JSON-like structured input for strictly defined compositions, a feature clearly aimed at programmatic generation and business pipelines.

The license split: open weights versus open source

For startups and developers building on BFL’s technology, understanding the licensing landscape of this release is critical. BFL has adopted a split strategy that separates the use of ‘hobbyist/researchers’ from ‘commercial infrastructure’.

1. FLUX.2 [klein] 4B: Released under Apache 2.0. This is a free software license that allows commercial use, modification and redistribution. If you’re building a paid app, a SaaS platform, or a game that integrates AI generation, you can use the 4B model royalty-free.

2. FLUX.2 [klein] 9B & [dev]: Released under the FLUX non-commercial license. These weights can be downloaded and experimented with by researchers and hobbyists, but cannot be used for commercial applications without a separate agreement.

This distinction positions the 4B model as a direct competitor to other open-weights models such as Stable Diffusion 3 Medium or SDXL, but with a more modern architecture and a permissive license that removes legal ambiguity for startups.

Ecosystem Integration: ComfyUI and more

BFL is clearly aware that a model is only as good as the tools used to run it. Coinciding with the model’s launch, the team released official workflow templates for Comfortablethe node-based interface that has become the standard integrated development environment (IDE) for AI artists.

The workflows (particularly image_flux2_klein_text_to_image.json and the editing variants) allow users to immediately drag and drop the new capabilities into existing pipelines.

The community response on social media focused on this workflow integration and its speed. In a post on

Why it matters for AI decision makers in enterprises

The release of FLUX.2 [klein] signals a maturation in the generative AI market, moving beyond the initial phase of novelty into a period defined by utility, integration and speed.

For Lead AI Engineers who are constantly juggling the need to balance speed and quality, this shift is crucial. These professionals, who manage the entire model lifecycle from data preparation to implementation, are often faced with the daily challenge of integrating rapidly evolving tools into existing workflows.

The availability of a distilled 4B model under an Apache 2.0 license provides a practical solution for those focused on rapid deployment and tuning to achieve specific business goals, allowing them to bypass the latency bottlenecks that typically plague high-fidelity image generation.

For Senior AI Engineers focused on orchestration and automation, the implications are equally significant. These experts are responsible for building scalable AI pipelines and maintaining model integrity across environments, often while working under strict budget constraints.

The lightweight nature of the [klein] family directly addresses the challenge of implementing efficient systems with limited resources. By using a model that fits within consumer-grade VRAM, orchestration specialists can design cost-effective, local inference pipelines that avoid the heavy operational costs associated with massive proprietary models.

Even for the IT security director, moving to capable, locally executable open-weight models offers a clear advantage. Tasked with protecting the organization from cyber threats and managing security operations with limited resources, reliance on external APIs for sensitive creative workflows can be a vulnerability.

A high-quality model that runs locally allows security leaders to punish AI tools that keep proprietary data within the corporate firewall, balancing the operational demands of the business with the robust security measures they must maintain.

#Black #Forest #Labs #launches #open #source #Flux.2 #klein #generate #images