Black Forest Labs Launches 'FLUX 3 Image' With Pixel-Preserving Edits and Bounding-Box Layout Control

On October 1, 2026, Black Forest Labs officially released 'FLUX 3 Image', an image generation and precision-editing model built on the multimodal FLUX 3 backbone. It combines text-to-image synthesis, image-to-image conversion, local editing, and multi-reference composition in a single model and API path, available through the BFL Playground and API.

Image source: Black Forest Labs (@bfl_ai)

The release is summarized by the company's tagline, "Control every pixel." Because regions outside the edited area stay unchanged across repeated revisions, the company positions it as a control-first model suited to iterative work such as advertising layouts and product imagery.

Bounding-Box Layout and Pixel-Preserving Multi-Turn Edits

The biggest technical change in FLUX 3 Image is bounding-box-based composition control.

Users place each element directly on the canvas by assigning a box plus a short description on a 0-to-1000 coordinate grid. The same approach works both when composing an image from scratch and when precisely editing part of an existing image. The company's launch demo shows a poster in which elements such as boots and text blocks each carry a labeled box.

• Pixel-preserving edits: in multi-turn partial edits, every pixel outside the user-specified region is kept unchanged. Users can recolor, move, replace, or remove one element at a time, reducing drift across the whole frame during iteration.
• Text and typography handling: the model supports in-place text editing within the selected region and typography rendering, and the company says lettering holds up in text-heavy layouts.

The launch materials do not include independent benchmark evidence that it outperforms the strongest rival image models, so the control interface itself — rather than a proven quality lead — is the point to watch in this release.

Up to 4K Output and Composition From Up to 10 Reference Images

Resolution and reference-composition range were expanded together.

Rendering spans from quick 768px drafts to native high-resolution output, with up to 4K output preserving detail, faces, and color, according to the company. Some third-party documents cite roughly 5456x3072 pixels as a 4K-class output example, but the official announcement wording is "up to 4K," which this article follows.

• Up to 10 references: up to ten different images — a product, a person, a room, a style — can be combined in a single request. References are cited in order within the request.
• Intended uses: advertising layouts, product imagery, and mockups requiring frequent revision are the primary targets, with identity and fine detail preserved across references, according to the company.

50% API Discount and the Commercial-Weights / Open-Weights Roadmap

Access and licensing come in three distinct tracks with different availability at launch, so they should be read separately.

Playground and API access started as of October 1, 2026. The API carries a limited-time 50% discount through October 8, 2026. Some third-party documents mention $0.0205 per image at 768px as the discounted entry price, which should be understood as the promotional price during the discount window. Pricing after the window needs separate confirmation.

• Commercial Weights: available to companies running image generation at scale. It is a license permitting fine-tuning and deployment on their own infrastructure, via direct inquiry to BFL.
• Open Weights: not released at the October 1 launch; a release in the coming weeks is planned. A free FLUX Tools precise-editing demo page was published separately.

Anyone evaluating adoption should first confirm the discount deadline (October 8) and the different availability status of the commercial and open weights.

Sources

• Black Forest Labs official model page: FLUX 3 Image
• Black Forest Labs official X (@bfl_ai): FLUX 3 Image launch thread
• BFL Playground: FLUX 3 Image playground