Adobe has pulled back the curtain on its latest batch of experimental technology at the annual Max conference, offering a glimpse into potential future features for its Creative Cloud apps and Firefly platform. Known internally as sneaks, the newly revealed projects explore advanced generative AI capabilities across video creation, 3D environment mapping, audio mixing, and photo retouching. While these tools remain strictly experimental with no guarantee of an eventual commercial release, they highlight Adobe's ongoing push to make complex creative workflows more interactive and intuitive.
Among the most striking demonstrations is Project Screen Test, a tool aimed at simplifying the development of AI-generated characters for video content. Instead of relying solely on static text prompts to define a character, creators can converse directly with the AI as if talking on a FaceTime call to establish a detailed personality, backstory, temperament, goals, flaws, fears, and movement quirks. Users can specify baseline traits like an awkward or shy temperament, select or describe a voice or upload reference audio, and then generate animated screen tests by placing the character into custom scenarios. Adobe senior principal research scientist Gautham Mysore notes that these interactive layers are designed to bridge the gap between creative intent and precise control, offering more consistent personalities than standard generative video tools.
Key takeaways
- Project Sight Span transforms flat photos, paintings, and illustrations into navigable 3D environments.
- Project Sound Bender lets users mix and sync audio using an interactive cursor canvas and text prompts.
- Project Rinse Repeat automatically adapts localized photo edits and product shadows to other subjects in an image.
- Project Screen Test allows creators to build character personalities and direct them through conversational interviews.
Another major highlight is Project Sight Span, which reframes standard photographs into fully explorable 3D spaces that you can navigate through and reposition. The tool generates background elements and objects outside the original frame that feel like natural parts of the image, allowing users to turn around and see what lies behind the subject. Although the initial rendering appears pixelated and ill-defined, the resolution quality restores automatically once a specific viewing angle is chosen. The feature preserves original artistic styles and keeps facial features and tattoos surprisingly recognizable, functioning much like an aggressive version of Apple's Spatial Reframing feature.
Audio production and standard retouching also received experimental upgrades during the showcase. Project Sound Bender utilizes text prompts, such as requesting bubbling sci-fi cauldron noises, to generate sound effects tied to video clips. It provides a square sound canvas where users can drag a cursor around corners assigned to distinct AI-generated sounds to adjust dense gurgling, rumbling boiling rocks, pitched bloops, or crackling electric fire in real time. Meanwhile, Project Rinse Repeat integrates directly with Photoshop to copy specific localized edits made to one person or object—such as airbrushing acne or adding marketing product shadows—and automatically adapts them to other subjects in the frame while keeping all modifications fully native and editable.
Comments