Black Forest Labs Launches FLUX 3 Image: Bounding-Box Edits Leave Pixels Outside the Box Unchanged

What Happened

On October 1, Black Forest Labs announced on its official X account @bfl_ai: "Introducing FLUX 3 Image. Control every pixel." This is the first productized release of the image branch after July's FLUX 3 multimodal architecture announcement (the Self-Flow framework spanning image, video and audio). The model ships through a unified API endpoint with two jobs: generate brand-new scenes from scratch, or modify existing images with text instructions plus spatial coordinates — with search grounding on by default to verify visual details (per third-party coverage).

The core feature is pixel-level control: every canvas maps to an aspect-ratio-independent 0-to-1000 normalized grid; each element gets an ID, a description and a bounding box [y_min, x_min, y_max, x_max]. Developers can hand in coordinates directly or have an LLM generate the layout table from a one-line brief; regions outside the boxes stay untouched across edit turns. BFL calls the editing "pixel perfect"; independent reviewers note that whether pixels outside the intended region stay unchanged beyond the company's samples needs testing on real files.

Key Facts

1. Availability: Hosted API and playground are live with a free demo; API is 50% off through Oct 8. Enterprises can license commercial weights to run and fine-tune on their own infrastructure.
2. Open weights: BFL has previously released hosted versions first and open weights later. The open-weights version of FLUX 3 Image is expected "in the coming weeks" with no firm date.
3. Specs: Native output up to 4K (including a 5,456 x 3,072 sample); up to 10 reference images per session; bounding-box layouts plus accurate multi-language text rendering. Early testers (Canva, Magnific, Krea, Picsart) report sharper detail and more coherent layouts than FLUX.2 Pro.
4. Rivals: Just before launch, Ideogram announced its own editing-focused 4.5, also planned for open weights.

Context

In multiple side-by-side comparisons, FLUX ranks ahead of Midjourney V7, DALL-E 3 and Imagen 3. The same day, Cloudflare released its open-source Clef decision models, which let agents make structured decisions without generating text.