Black Forest Labs launched FLUX 3 Image, a high-precision AI image-generation and editing model built on the FLUX 3 foundation, designed for controlled composition, multi-step edits, and 4K output.

The system uses bounding-box based control to specify where subjects appear, preserving unedited regions with bit-identical fidelity and enabling edits that target only selected areas.

FLUX 3 Image supports text-to-image, image-to-image, text rendering, and photorealism, with the ability to perform multi-step edits while maintaining layout and preserving untouched pixels.

A two-part layout approach is used: a global scene caption and an element table in JSON listing each element with an id, bounding box, and description; captions reference elements by id.

Users can compose scenes by drawing up to ten bounding boxes, include as many as ten reference images, and output resolutions up to 4K.

A free demo is available for testing the tool.

Local deployment options are discussed, with commercial weights suitable for private infrastructure and careful cost considerations for hosted versus self-hosted use.

Launch pricing includes discounts across multiple output sizes, with refunds not charged; total production spend should consider revisions and preservation costs.

The story is published on October 2, 2026, as part of a tech news outlet covering AI developments.

Flux 3 Image is available via BFL Playground and API as of early October 2026, with a 50% introductory discount through October 8, and an open-weight version expected shortly.

There are separate access tiers: commercial weights via BFL and a forthcoming open-weight version, with launch pricing running a limited-time discount from October 1 through October 8, 2026.

API access is discounted 50% through October 8, with on-premise licensing and fine-tuning options; an open-weight release is anticipated in the coming weeks.