Adobe Max Sneaks Preview 3D Scene Building, Conversational AI Characters and Copy-Paste Photo Editing
Adobe's Max Sneaks showcase includes experimental tools that turn images into explorable 3D scenes, design AI characters through conversation, and turn photo editing into a copy-and-paste operation. The Verge reported the lineup on 11 November 2026. Sneaks are Adobe's experimental showcase format, and the projects below are prototypes presented as experiments.
The most visual of the set is Project Sight Span. It takes photographs and other images and builds a 3D environment that users can navigate through and reposition. According to the report, it works with any flat image, including photos, paintings and illustrations, and it can preserve and replicate the artistic style of the source. It also generates objects and backgrounds that sit outside the original frame, and resolution quality is restored once the user selects the chosen angle. That last detail matters for anyone who has tried to push a cropped image past its pixel budget. The tool appears to treat the original frame as a starting constraint rather than a boundary.
Project Screen Test targets a narrower problem: making AI-generated characters easier to develop and use in video content. The workflow starts with describing base personality traits and assigning a voice, which can come from a preset, a text description or uploaded sample audio. Users then expand the persona by talking with the character in a FaceTime-style conversation, outlining backstory, temperament, goals, flaws, fears and movement quirks. The interface is built around an audition rather than a prompt box, so the user directs the character through dialogue instead of only through text instructions. From there, the tool can generate animated screen tests of the character in a described environment or scenario.
Project Sound Bender is an AI audio-mixing feature for customising sound effects on a square sound canvas. The process begins with uploading a video clip and describing the desired audio in a text prompt. The tool then generates sound effects and syncs them to the video automatically.
The showcase also includes a tool that turns photo editing into a copy-and-paste process.
The broader context here is a shift in how generative systems are steered. Most mainstream image and video generators still rely on a single prompt and a re-roll when the result misses. Screen Test instead treats a character as something that accumulates traits over a session, closer to casting and rehearsal than to prompt engineering. Sight Span similarly moves from a single output to a navigable scene that can be re-framed after generation. For practitioners, the open question is less whether these demos look convincing and more what the underlying controls look like: how persona state is stored, how consistency holds across a long conversation, and how spatial reconstruction from a single flat image handles occluded geometry.
In my view, the Sneaks format is a useful signal of where a vendor's research is pointed, but it is not a product commitment. Experimental demonstrations of this kind often differ substantially from what reaches a shipping application, and the gap tends to be widest where the demo depends on controlled inputs. Developers and creative teams evaluating these directions should treat them as research previews until the tools are generally available and their behaviour can be tested on their own material.
Even so, the direction is clear enough to be worth tracking. If the 3D scene tool holds up, a single photograph becomes a starting point for camera moves and layout work that previously required a 3D artist. If the character tool holds up, pre-production for short video could use characters that have been rehearsed, not merely generated. Sound Bender suggests that foley and effects work, which often consumes hours of an editor's time, could be drafted directly from a clip and a sentence of description. Each of these would shorten a specific bottleneck in creative production, and each would still need human judgement wherever the output has to match intent.
