Black Forest Labs Introduces FLUX 3 Image with Bounding Box-Based Generation and Editing

Black Forest Labs has introduced FLUX 3 Image, the image generation and editing component of its multimodal model suite. The model distinguishes itself through bounding box-based composition, allowing users to define layout coordinates on a 0–1000 grid to control element placement precisely.

Layout and Agent Capabilities

The system uses a structured prompt format combining a global caption with a JSON element table that references specific bounding boxes. An integrated agent feature can automatically generate these layout prompts from a single text line and aspect ratio, while a prompt upsampler converts short requests into dense captions without altering user-defined box coordinates.

Editing and Reference Features

FLUX 3 Image supports pixel-perfect editing, where modifications to specific bounding boxes leave the rest of the image unchanged. Users can also utilize up to 10 reference images to guide the generation of complex designs, reducing the effort required for detailed visual outputs.

Resolution and Access

The model supports native 4K rendering, preserving fine details in textures, faces, and colors. Examples include images rendered at 5456 × 3072 pixels. Access is available via the Playground at flux-tools.bfl.ai/precise-editing or through the BFL API. Commercial weights licenses are provided for self-hosted fine-tuning and deployment.