Adobe MAX Sneaks 2026: Ten AI Prototypes That Could Reshape Photoshop and Premiere

Adobe used its MAX Sneaks 2026 showcase to lift the curtain on ten artificial intelligence research projects that hint at the future direction of its creative software. Presented on November 11, 2026, and hosted by actor Terry Crews, the demos ranged from applying a single edit consistently across every shot in a scene to physically grabbing elements inside an image or directing a video while it plays. Crucially, these are prototypes rather than shipping products — but for the first time, Adobe is letting the public get hands-on through a new Labs Playground.

  • Ten AI prototypes: Adobe’s MAX Sneaks 2026 featured research spanning video, audio, imaging and design workflows.
  • Public testing opens up: A new Labs Playground lets the community try selected projects, including early versions of Deep Brush and Hands On.
  • No launch dates: Adobe has not announced pricing, release windows or the training data behind these systems.

Video and audio: editing entire scenes and rescuing sound

The video-focused block centers on four projects. Project Scene Sweep understands characters, locations and scenes across an entire sequence, so a single editing decision can be propagated through every cut and camera angle. Project Live Set turns live video into a responsive canvas: users can swap a subject, transform an environment or reframe the camera while the footage keeps rolling. Project Screen Test lets creators develop a fictional character’s look, voice and personality, then converse with that character in natural language. Project On Cue gives precise control over when an action occurs in generated video, using a timing signal that can be drawn by hand or extracted from music.

On the audio side, Project Sound Bender lets users describe a sound and isolate it from an already-finished mix, then explore variations on an interactive canvas to shape it in real time. The underlying logic is classic Adobe: fold scattered capabilities into a single, cohesive workflow.

Imaging and design: brushes that understand, edits you can grab

Five projects target image creation and design. Project Deep Brush uses AI to comprehend an image’s shape, color and lighting, so brushstrokes follow contours, color changes preserve shadows, and light can be painted directly into the scene. Project Hands On combines object segmentation with vision-language models to create on-canvas controls, letting users manipulate elements by hand. Project Rinse Repeat captures edits made in one region of an image and applies them elsewhere with adaptation — and because the results are rebuilt with native Photoshop tools, they remain editable.

Rounding out the group are Project Sight Span, which converts a still image into a navigable environment for finding angles and framing beyond the original shot, and Project Link Up, which interprets creator gestures: drag a car into an illustrated parking space and the project adapts its perspective, scale and style, returning an editable vector. These are the laboratory experiments that eventually land in products — much like Adobe’s recent push to embed AI agents into Photoshop, Premiere and Firefly.

Why the Sneaks are touchable this year

Adobe wants real feedback before deciding what becomes a product, and the Labs Playground is how it plans to get it. For the first time, a selection of these research projects can be tested: early builds of Project Deep Brush and Project Hands On are available, alongside Project Custom Filters, a capability Adobe is exploring for Photoshop. The company is soliciting opinions at labs@adobe.com and framing the effort under its motto, “Dream big, create bigger.”

Opening the lab to the public signals that not everything dazzling on stage becomes a product — and that feedback decides.

That openness connects to a broader corporate strategy: the same approach that brought Adobe’s tools to Google and Microsoft and placed rival AI inside Photoshop. Against standalone utilities, Adobe’s bet remains keeping everything inside one workflow.

The honest reading is that Sneaks remains a dazzling show best filtered through reality. Many of these ideas are genuinely useful — editing a scene once, or recovering a sound from a closed mix, could save hours of real work — and opening the Playground is a healthy shift. But it is also the familiar trap: magic is shown without the fine print. There are no prices, no dates, no named models, and not a word about what data trained these systems that “understand” an image or clone a character’s voice. That question — where the material comes from — hangs over all creative AI, and it is one Adobe has begun addressing elsewhere by training models on Disney’s characters under Disney’s own control. So note what is coming: Scene Sweep and Hands On look like the future of Photoshop and Premiere. But until they have a price, a date and a clear answer about their data, they are exactly what their name says: a preview.