Black Forest Labs open-sources FLUX 3 Klein as internal tests top Qwen, while early users push back

News Editor

2026-10-16 08:22:53

German AI company Black Forest Labs has open-sourced its image model FLUX 3 Image Klein, introducing a release that combines a 5.6 billion-parameter image generation component with a 4 billion-parameter Qwen3-VL module for prompt understanding, bringing the effective total to about 9.6 billion parameters. The model supports native 4K image generation, image editing, and up to 10 reference images. After quantization, it can run on RTX GPUs with as little as 8GB of VRAM, while an RTX 5090 can generate a one-megapixel image in about four seconds. Black Forest Labs said its internal evaluations showed FLUX 3 Klein outperforming Qwen-Image 2.1 in both text-to-image generation and image editing, with gains of 91 Elo and 70 Elo over the previous FLUX.2 Klein 9B in those two categories. But early Reddit testing painted a weaker picture, with users sharing failed attempts at pose adjustment and viewpoint switching, and some saying results trailed the earlier 4B model. In a separate third-party blind test covering 200 real editing tasks, FLUX 3 Klein scored 6.93, below Nano Banana 2.1 at 7.20, though its per-image cost was only one-fifth as high.

German AI company Black Forest Labs has open-sourced FLUX 3 Image Klein, a new image model that the company lists at 5.6 billion parameters. That figure covers only the image generation portion. The full setup also relies on the 4 billion-parameter Qwen3-VL model to interpret text prompts, bringing the combined total to roughly 9.6 billion parameters.

The model supports native 4K image generation, image editing, and as many as 10 reference images. After quantization, it can run on RTX GPUs with as little as 8GB of VRAM. On an RTX 5090, generating a one-megapixel image takes about four seconds.

Internal benchmarks and user tests diverge

According to Black Forest Labs, internal evaluations showed FLUX 3 Klein outperforming Qwen-Image 2.1 in both text-to-image generation and image editing. Against its own previous FLUX.2 Klein 9B, the new release improved by 91 Elo and 70 Elo in those two categories.

Early hands-on feedback from Reddit was less favorable. Users posted cases in which the model failed to adjust human poses or switch viewing angles, and some said the results looked worse than those from the earlier 4B model.

In a separate third-party blind test covering 200 real editing tasks, FLUX 3 Klein scored 6.93, slightly below Nano Banana 2.1 at 7.20. Still, the cost per image was only one-fifth of Nano Banana 2.1. That same test found the model prone to errors in pose adjustment and multi-image composition.

Company recommends prompt expansion

Black Forest Labs said users are better off expanding prompts with an additional language model before running the system, warning that short instructions can lead to mediocre results. To address that, the team also released a 9 billion-parameter prompt expander.

License terms and API pricing

FLUX 3 Klein is released under the FLUX Kommunity License. Individuals and companies with combined annual revenue below $5 million can use it for commercial image generation under the license terms, though commercial fine-tuning requires separate authorization.

API pricing starts at $0.009 for a 1K image.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.

200

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.