Flux 2 Pro

FLUX.2 is Black Forest Labs' flagship AI image generation and editing model, released in late 2025. The Pro variant targets production use: up to 4 megapixels, multi-reference control (up to 10 reference images) for character and style consistency, and strong editing capabilities. The model uses latent flow matching with a Mistral-3 24B vision-language model and a rectified flow transformer, enabling fast generation and strong prompt adherence.

FLUX.2 Pro suits creators who need high-fidelity images with precise control—hero marketing frames, character series, and layouts where typography and composition matter. Black Forest Labs cites roughly 60% accuracy on complex typography on first attempt for many infographic-style prompts.

Black Forest Labs provides FLUX.2 in multiple variants: klein for real-time use, pro for production scale, flex for fine-grained control, and max for highest quality with grounding. Product teams pick the variant that matches latency, control, and licensing for their stack.

FLUX.2 generates and edits images at up to 4 megapixels with photorealistic quality, precise color matching, and improved text rendering. You get production-ready assets for print and high-resolution digital use. When you need images that will be viewed at full size or printed, the 4MP ceiling gives you headroom for cropping and composition.

The model can process up to 10 reference images at once. The first six get high-fidelity treatment for character identity, product detail, and style; images 7–10 act as supplementary references. That makes it well suited for serial content and brand consistency—same character or product across many images.

Edit images through text prompts with multi-reference support, color control via hex codes, pose guidance, and the ability to extract and recompose elements. FLUX.2 works as both a generator and an editor in one pipeline, so you can iterate on an image without leaving the model.

FLUX.2 achieves roughly 60% accuracy on complex typography on first attempt, making it suitable for infographics, UI mockups, and marketing materials that include text. While not as specialized as some text-first models, it is capable enough for many captions, labels, and headlines.

The model uses latent flow matching rather than traditional diffusion, combining a Mistral-3 24B vision-language model with a rectified flow transformer. This enables faster generation times and strong prompt adherence.