Picsart adds Flux 3 Image with box-based layout control and native 4K

One model, three jobs

Flux 3 Image is the image part of Flux 3, which Picsart describes as Black Forest Labs' multimodal model family for video, audio, images and actions. The model is natively trained to understand image layout and composition.

There is no mode to switch, because the inputs decide the task:

• A prompt on its own generates a new image from a description.
• A prompt plus one image edits that image, from a single detail to its whole style.
• A prompt plus 2 to 10 images restyles them or builds a new composition from all of them.

Picsart says prompt following is strong enough that a single line can produce a well-composed image, and that the output is full resolution and ready for ads, posters and brand assets.

Layout with bounding boxes, edits that leave the rest alone

Layout control is described as the signature feature. Whatever aspect ratio is chosen, the canvas works as a grid running from 0 to 1,000. Each important element gets its own box with a short description, and one line ties the scene together. The model then renders every element inside its box. The boxes are optional, and a plain written prompt works too.

For editing, several targeted changes can happen at once, such as re-describing, replacing or moving an element, while everything untouched stays where it was. Picsart says the edits hold together across successive rounds, so an image can be refined step by step.

In one Black Forest Labs demo, a black and white surf photo gets two changes in a single pass: the black wetsuit becomes a bright red neoprene wetsuit and the white surfboard turns bright red, while the cresting wave and cloudy sky stay as they were. In another demo, two divers are added to a photo while the original around them is preserved.

Up to 10 references and native 2K and 4K

The model accepts up to 10 reference images. Each gets a token in the order it was added, starting at ref_image_0, and the prompt cites them in one line. Flux 3 Image decides where each reference sits and how big it is before rendering the frame. A streetwear demo combines six references (a vest, a tee, jeans, a duffel bag, a beanie and sneakers) into a Times Square fashion portrait.

Output is rendered natively in 2K or 4K, which Picsart says preserves textures, faces and colors. A soba shop demo comes out at 5456 x 3072 pixels, or 16.8 megapixels. Small lettering is also said to stay sharp, which suits posters, magazine covers, editorial spreads and long infographic layouts.

Availability and who it targets

Flux 3 Image is integrated into Picsart AI Playground, which works on the web and in the desktop app. It can also be reached through the Picsart CLI, MCP, REST API and SDK for those building it into their own projects. Generations use the existing Picsart credit balance, with no third-party API keys or separate subscriptions. The source gives no pricing details.

Picsart points to four groups of users. Marketers and ad teams can make several changes in one request across repeated review rounds. Designers can place headlines, hero images and supporting details with boxes. Educators and explainer creators can rely on sharp small text in infographics. Fashion and ecommerce brands can blend up to 10 references while keeping products, characters and styles recognizable.