German AI startup Black Forest Labs, founded by the original creators of Stable Diffusion, has launched FLUX.2, a new family of AI models for generating and editing images. According to the company, the system is designed for professional creative workflows, aiming for reliability and control rather than one-off demonstrations.
FLUX.2 introduces several key enhancements over its predecessor. The models are built to maintain character and style consistency while referencing up to ten images simultaneously. Black Forest Labs also states that the system shows significant improvements in generating complex and legible text, a common challenge for image models. It can generate and edit images at resolutions up to 4 megapixels while preserving fine details.
An ‘Open-Core’ Family of Models
The launch follows Black Forest Labs’ “open-core” strategy, which combines commercial products with open-weight models for the developer community. The FLUX.2 family includes several variants:
- FLUX.2 [pro] and [flex]: These are commercial models available through an API. The [pro] version is optimized for quality and speed, while [flex] gives developers more control over parameters to balance speed and detail.
- FLUX.2 [dev]: This is an “open-weight” model, meaning its files can be downloaded and run on local hardware. While available to the public, it requires a commercial license for business use.
- FLUX.2 [klein]: An upcoming smaller model that will be fully open-source under an Apache 2.0 license.
- FLUX.2 – VAE: This foundational component handles image compression and is also fully open-source. The company says this allows businesses to integrate it into their own systems, ensuring interoperability without being locked into a single vendor.
According to benchmarks published by the company, FLUX.2 [dev] outperforms other available open-weight models in text-to-image generation and editing tasks. A VentureBeat article highlights a cost comparison, noting that the FLUX.2 [pro] model is priced significantly lower per image than competing high-end models like Google’s Gemini 3 Pro Image Preview. For example, a standard 1-megapixel image from FLUX.2 [pro] costs about $0.03, while a similar image from the competitor could cost more than four times that amount.
The FLUX.2 system is built on a technical architecture that combines a large language model from Mistral with a transformer, which helps it better understand prompts and spatial relationships. By releasing a mix of proprietary and open models, Black Forest Labs aims to foster community innovation while providing scalable, production-ready tools for businesses.
Sources: Black Forest Labs, VentureBeat
