Adobe showcases four AI prototypes at Max: 3D from photos, a sound square, and copy-and-paste edits

Every year at its Max conference, Adobe showcases “sneaks”: experimental prototypes that may or may not become features in Creative Cloud or Firefly. This year, all the projects share a theme: generative AI is moving beyond a prompt box to become a set of intuitive controls. Here are four of the most notable projects.

Characters and 3D from a single image

Project Screen Test tackles a familiar frustration with AI video: a character comes out slightly different every time, and defining its personality through text is nearly impossible. The proposed approach is to “audition” the character, as you would an actor. You set basic traits such as “anxious, shy, awkward,” then choose a voice from presets, describe it in words, or upload an audio sample.

Next, you talk to the character as if on FaceTime, asking about its background, temperament, goals, fears, and manner of moving. Everything discussed feeds into a Personality tab, where you can adjust traits manually. You then describe a scene and get an animated screen test.

For now, this is a standalone creative experiment, and how it fits into Adobe’s video tools remains unclear. Still, according to The Verge’s writer, the resulting personality is noticeably more distinctive and consistent than that of typical “soulless” AI characters.

Project Sight Span takes any flat image—a photo, painting, or illustration—and builds a 3D space around it. You can rotate the camera and even see what lies “behind” the original viewpoint: a neural network fills in objects and backgrounds beyond the original image. The 3D view itself is pixelated and blurry, but once you choose an angle, image quality is restored. The artwork’s style is preserved, while tattoos and facial features remain recognizable after the camera position changes. The writer compares it to Apple’s Spatial Reframing “on steroids” and notes that the result is already hard to call a photograph.

A sound square and copy-and-paste edits

Project Sound Bender offers sound mixing without tracks or equalizers. You upload a video clip and describe the sound you want, such as a “bubbling sci-fi cauldron.” The tool generates effects and automatically synchronizes them with the video. A square “sound canvas” then appears, with a different sound variation in each corner: thick bubbling, the rumble of boiling rocks, high-pitched “bloops,” or the crackle of electrical fire, for example. Moving the cursor around the square blends the variations in real time, and you can reassign the corners.

Project Rinse Repeat will appeal to anyone doing retouching at scale. You carefully remove blemishes from one person’s face once, and the tool applies equivalent edits to everyone else in the group shot. Another use case: draw a shadow beneath one product in an advertising graphic, and matching shadows appear beneath the others, even if their shapes and sizes differ. The key engineering decision is that Rinse Repeat does not use a neural network to “redraw pixels”; it recreates the edits using Photoshop’s native tools. The result therefore remains in editable layers.

What this means

First, do not expect these features to arrive in a product tomorrow. Adobe explicitly says sneaks are technology previews, with no guarantee of release. But almost all the projects build on the company’s generative AI work, and AI is now central to its growth strategy, so some of these ideas will almost certainly reach users.

Second, product managers and developers can learn from the interface patterns themselves. A conversation instead of a lengthy character description, a two-dimensional plane instead of a bank of sliders, and “repeat my edit” instead of another round of manual work. These are attempts to give users back the control they lose when working with a single prompt. Adobe senior research scientist Gautham Mysore put it this way: intuitive interaction is also about helping experienced creators control results more precisely, not just making things easier for newcomers.

Third, there is a warning sign for the market. If tools like these become useful enough, they could eventually replace established software and stock libraries used by creative professionals. Some features look like toys for now, but they have practical uses. There are limitations: the demos were seen only through descriptions and images, the intermediate 3D view is low-quality, and Screen Test is not yet integrated into any workflow. If you build creative AI tools, treat these prototypes as a cheat sheet for interfaces rather than production-ready technology.