Black Forest Labs drops FLUX.2 [max]

Black Forest Labs just launched FLUX.2 [max], billing it as its highest-quality image model to date.

Web grounding and 10 reference images

The standout addition is grounded generation. Instead of hallucinating blindly on fresh topics, the model can search the web for real-time context before crafting an image.

It also now accepts up to 10 reference images. This is built specifically to keep products, recurring characters, and overall aesthetic styles consistent across multiple generations and edits.

According to Black Forest Labs, FLUX.2 [max] currently sits at number two on Artificial Analysis leaderboards for both text-to-image and image editing.

Why it matters

Asset consistency has always been the primary barrier preventing image models from taking over professional pipelines. Pushing the reference limit to 10 images directly targets brand assets and character drift, which should make repetitive commercial work far less frustrating.

Adding web search to the generation loop is equally pragmatic. Tying visual output to live context prevents the model from relying entirely on stale training weights when rendering timely subjects.

We think this is a compelling release for designers managing strict visual identities, though practical adoption will ultimately hinge on API pricing and deployment requirements that have yet to be detailed.