Black Forest Labs launched FLUX 3 Image with controls for placing and changing individual objects. The company says repeated edits preserve areas the user leaves untouched. Hosted access has begun, while downloadable weights remain planned for later. The method combines natural-language instructions with object identifiers and position boxes, giving image editing a more explicit layout mechanism.
Black Forest Labs, the developer of the FLUX image-model family, launched FLUX 3 Image for generating and editing pictures. The company says successive changes preserve pixels outside the requested edits. The release focuses on modifying individual parts of an existing composition, with preservation presented through company demonstrations rather than an independent measurement.[1], [2]
Users can place an object by drawing a bounding box on a visual canvas. Through the programming interface, positions can instead be supplied as structured descriptions. Each element receives an identifier, a position box and a description; a global caption refers to those identifiers. Ordinary natural-language prompting remains available without drawing boxes for every object.[1]
An edit can change an element’s description, move it or replace it. Reference images supply objects or appearance information, and their elements can also receive placement instructions. One provider demonstration changes a surfer’s wetsuit and board while retaining the wave and clouds.[1]
The model is accessible through a hosted service and an API. Black Forest Labs says downloadable weights are planned for the coming weeks. Commercial weight licensing is described separately from that planned release. The announcement therefore establishes hosted access while the distribution of open weights remains a later step.[1]
