Disclose AI imagery without labelling files by hand

Dreem can now add the official EU AI-content icon to the bottom-right corner of exported images. Digital Twin exports carry it by default. Workspace admins can turn it on for all other AI images in workspace settings, or off for Digital Twin. The label is added on download and Shopify push, and the stored asset stays unlabelled. Assets delivered through the Creative Force connector are not labelled. The label supports disclosure but does not by itself make an asset legally compliant.

Change course without waiting it out

There is now a stop button in the prompt box. It cancels a generation or a chat reply that is still running. After you stop, the card reads "Generation stopped" or "Response stopped", with a Try again button. This works in Dreem Studio on the web. Stopping inside Image Editing is not included yet.

The right input lands in the right position

When a generation starts from a Creative Force task, the connector now reads the source tags Creative Force already puts on each asset, such as front, back, or detail. Dreem uses them to place each input in the right position. You can see the view type Dreem received on each input and change it if it is wrong. Tags are optional. If an asset has no tag Dreem recognizes, it works as before.

Learn more

It reports only what it made, and asks which garment you mean

After a generation, the agent no longer says an output looks good or is ready to publish when that was never checked. It also no longer describes colors or patterns that are not in the image. The message now states what was produced. Separately, when an input photo shows more than one garment, such as a top and a bottom, the agent now confirms which one you want before it generates.

Upload a styling item once and reuse it

You can now upload your own styling images into the Outfit Builder wardrobe, arrange them, and filter the wardrobe to find them again. Saved items are then ready to use when you build later outfits. This is in Dreem Studio on the web.

Learn more

More outputs that match what you uploaded

This release carries quality fixes across Product Shot and Virtual Model. In Product Shot, outputs match the input more closely, with fewer cases of the wrong garment type, pattern, or color, or a shot tied to the wrong product. In Virtual Model, details from an input photo or an aesthetic reference, such as tattoos, accessories, or clothing, carry over onto the talent less often. Talent looks more consistent across a batch, with fewer anatomy problems such as extra arms or legs. Backgrounds and scenes are cleaner too. The quality work continues, and more fixes follow in the October 13 release.

The ones a customer may have already hit

In Creative Force connected sessions, a rejection or instruction with no markings showed a low-resolution, blurry image. A download triggered from Creative Force could also fail when the same file was sent twice. In Image Editing, the Edit tab could stay on loading or show an old version, and the timeline did not update until you reloaded the page. Choosing an earlier version of a regenerated video did not switch the player to it. Adding an unsupported file, such as a PDF, in Asset Hub gave no error message. Canvas tiles could sit unevenly in the grid. Creating a synthetic talent could leave out the talent type. All of these are fixed.

Clearer errors and easier version tracking

Error messages across Dreem now follow one standard format, so a failure reads the same way wherever it happens. The full screen viewer now shows which version of an asset you are looking at. Uploads in the Universal Overlay now check that you add no more than 10 files at a time.