Black Forest Labs Launches FLUX 3 Image Klein, a 5.6B Open Model

Black Forest Labs (BFL) has released FLUX 3 Image Klein, a new open-weights text-to-image model that sets a high bar for quality while being small enough to run on consumer hardware. This release marks a significant step in making high-fidelity AI image generation and editing accessible for local deployment and fine-tuning.

What is FLUX 3 Image Klein?

According to the company's announcement, FLUX 3 Image Klein is a 5.6 billion parameter model for text-to-image generation, image editing, and multi-reference generation using up to ten images. It shares core capabilities with the larger FLUX 3 Image model, including native 4K generation, pixel-perfect editing, and bounding-box controls for precise element placement. BFL states the model was trained using learnings from across images, video, and audio domains, which helps its outputs feel more grounded in real-world physics and interactions. The release includes both a base model for fine-tuning and a step-distilled version for low-latency generation, with builds optimized for NVIDIA hardware.

How does it perform against other models?

In BFL's internal evaluations, FLUX 3 Image Klein reportedly ranked first in text-to-image, multi-reference generation, and image editing among open-weights models, surpassing larger competitors. The company claims it also rated above closed models like Ideogram V4.5 and FLUX.2 [pro] in their comparison. At 5.6 billion parameters, it is notably smaller than other leading open models like Qwen-Image 2.1 (7B) and Krea 2 (12.9B). Compared to its predecessor, FLUX.2 [klein] (9B), the new model is said to have gained 91 Elo points in text-to-image and 70 in image editing while using 38% fewer parameters.

Why does its smaller size matter?

The reduced parameter count is a key feature for practical deployment. BFL highlights that the model was optimized for consumer GPUs, making it feasible to run locally and cheaper to adapt for specific use cases. The generator weights require 11.2 GB in BF16 format and 5.6 GB in FP8. With an INT8 build, the company says FLUX 3 Image Klein can run on NVIDIA GeForce RTX GPUs with as little as 8GB of VRAM through tools like ComfyUI. This opens the door for on-device image editing on laptops and allows teams with data sovereignty requirements to host the model on their own infrastructure.

What are the speed and licensing details?

BFL's data, measured by NVIDIA, shows the distilled model with an INT8 build can generate a 1-megapixel image in about 4 seconds on an RTX 5090 and under 9 seconds on an RTX 4070 via ComfyUI. Through the BFL API, the model is said to deliver lower latency than several competitors while maintaining high quality. The weights are released under the FLUX Kommunity License, which permits free commercial use for individuals and companies with annual revenue under $5 million, alongside research use. Larger organizations can obtain commercial licenses for self-hosting or use the paid API, which starts at $0.009 per 1,000 images.

How does this relate to creative work in Grafy?

For creators using Grafy's multimodal workspace, the advancement of efficient, high-quality open models like FLUX 3 Image Klein is directly relevant. In Grafy Creator, users can already generate and edit images using a variety of cloud models. The principles behind this release—such as multi-reference generation, bounding-box control, and pixel-perfect editing—align with features explored in our guides on multi-reference image editing and consistent AI characters. While Grafy does not host this specific model locally, its architecture supports integrating leading cloud APIs. Users interested in the technical details or considering fine-tuning such a model can use Grafy Reader to analyze the research paper or Grafy Chat to discuss its implications with different AI models.

Frequently asked questions

What is the main advantage of FLUX 3 Image Klein?

Its primary advantage is leading-edge image quality and editing capabilities from a relatively small 5.6B-parameter model. This combination of high benchmark scores and efficiency makes it practical for local deployment on consumer GPUs, reducing costs and enabling data-sovereign use cases.

Can I run it on a laptop with 8GB of VRAM?

Yes, according to BFL. Using the INT8 build of the model with compatible software like ComfyUI, it is designed to run on NVIDIA GeForce RTX GPUs with as little as 8GB of VRAM. This brings high-quality AI image generation and editing to more personal hardware.

What is the license for commercial use?

The model is released under the FLUX Kommunity License. It allows free commercial use for individuals and companies with under $5 million in annual revenue. Larger enterprises need to obtain a separate self-hosted license or use the commercial BFL API, which has its own pricing.

---

Source: Hacker News — front page — FLUX 3 Image Klein