Black Forest Labs launches Flux 3 Image with multi-step editing and pixel preservation

Black Forest Labs launched Flux 3 Image on October 1, 2026, introducing a structured JSON bounding-box composition system and bit-identical pixel preservation to eliminate visual drift in multi-step AI image editing. The model supports native 4K output and programmatic agentic workflows. API access is available with a 50 percent discount through October 8, 2026, starting at $0.0205 per image, with an open-weight release expected in the coming weeks alongside competitor Ideogram's upcoming open-weight release.

Flux 3 model releases

• ▪Black Forest Labs released Flux 3 Video on August 4, 2026, as part of its multimodal Flux 3 family.
• ▪Black Forest Labs launched Flux 3 Image, the image-focused model of its multimodal Flux 3 family, on October 1, 2026.

Flux 3 image capabilities

• ▪Flux 3 Image accepts up to ten reference images per generation, which can be supplied as URLs or base64-encoded data.
• ▪Flux 3 Image supports text-to-image, image-to-image, text rendering, photorealism, and native output up to 4K resolution.

Flux 3 Image pricing

• ▪Through October 8, 2026, Black Forest Labs offers a 50 percent introductory discount on Flux 3 Image API, pricing entry-level images at $0.0205 per image at 768px.
• ▪The price of a 1K-resolution image on the Flux 3 Image API will increase to approximately $0.048 after the introductory discount window closes.

Structured layout control

• ▪The bounding-box grid coordinates in Flux 3 Image are decoupled from pixel resolution and aspect ratio, allowing the same layout to apply consistently across different output sizes.
• ▪Flux 3 Image accepts a structured JSON array of scene elements where each element has a unique identifier, a natural-language description, and coordinates mapping to a normalized 0-to-1000 grid.
• ▪The stable JSON element IDs and coordinates in Flux 3 Image allow Large Language Model agents to treat the element table as a persistent project file to perform iterative, targeted edits.
• ▪Flux 3 Image includes a prompt upsampler that expands short instructions, but stable box IDs and coordinate arrays bypass this rewrite step to reach the model unchanged.

Pixel preservation during editing

• ▪Black Forest Labs' published measurements show that a cat-removal operation left 80.7 percent of pixels identical, a bird-addition left 86.4 percent unchanged, and a hair-recolor left 67.8 percent unchanged.
• ▪Flux 3 Image preserves pixels outside a selected bounding box with bit-identical numerical fidelity during multi-step editing, preventing visual drift across successive edits.

Upcoming open-weight releases

• ▪An open-weight version of Flux 3 Image is expected to be released in the coming weeks following the initial API launch.
• ▪Ideogram announced its own editing-focused model, Ideogram 4.5 Precise Edit, which is also scheduled to ship as open weights soon.

Debatable claims

• ▪AI developers are justified in delaying open-weight releases behind commercial API launches
• ▪AI image editing platforms should prioritize programmatic JSON layouts over natural-language instructions
• ▪Bit-identical pixel preservation is superior to contextual inpainting for professional AI image editing
• ▪Pixel-preservation editing is now a baseline standard rather than a key competitive differentiator for AI image generators