The Creative Park at Adobe MAX 2026 in Miami Beach.
Adobe has shown 10 research projects at its MAX Sneaks session in Miami Beach. Most of them let creators edit with a brush, a drag or a gesture instead of a typed prompt. Anyone can already try two of them in a new Adobe Labs Playground.
Actor Terry Crews hosted the session on Wednesday evening. Adobe listed all 10 projects in a blog post by John Yang , covering video, audio, imaging and design. Sneaks are research previews, not products, and Adobe has not said which of them will ship or when. The company says the work shown at Sneaks feeds its next generation of tools.
Adobe Labs explains the thinking in a research piece, Beyond the prompt . Adobe researchers Victoria Hollis and Kimberly Welchons wrote it.
“Chat is the most common interface for generative AI, but it is insufficient for precise visual work,” the researchers wrote.
It draws on eight studies with 82 working and aspiring creative professionals, run from 2024 to 2026. The authors also reviewed 27 research papers. The authors say creatives find their intent by making things, not by describing them up front. They add that the sample is not representative of all creative workers and leans towards visual disciplines.
Four of the projects deal with video. Project Scene Sweep applies one edit across a multi-shot sequence. The change then holds across cuts and camera angles. Adobe says the team trained its own editing model for multi-shot video.
Project Live Set lets a creator change a subject, the setting or the camera while the video plays. Project On Cue controls when something happens in a generated clip, such as a flash of light. A creator draws its timing signal by hand or takes it from a video or a piece of music.
Project Screen Test builds a fictional character’s look, voice and personality. The creator can then talk with the character to test how it might react in different scenes.
Project Sound Bender is the only audio project. A creator describes a sound, and the tool pulls it out of a finished mix. Variations then appear on an interactive canvas. There, the creator can shape the sound live or match it to movement in a video.
The image and design projects lean hardest on direct editing. Project Deep Brush splits a photo into layers of shape, colour and lighting. Brushstrokes can then follow the contours of a surface. A creator can also paint light into a scene and keep its natural shadows.
Project Hands On analyses an image and builds controls on the canvas, such as control points on a mug’s handle. A creator reshapes the detail by dragging, not by rewriting a prompt. Adobe says it grew out of the Edit object tool in Firefly. At first, the AI models were too slow for it.
Project Rinse Repeat copies edits from one part of an image to another and adapts them to the new area. It rebuilds them with native Photoshop tools, so they stay editable. Project Link Up turns a drag into an instruction. Drop a car into an illustrated parking space, and it adjusts perspective, scale and style. The result stays editable vector artwork. Project Sight Span turns a single image into a space a creator can step into and frame new shots from.
The Adobe Labs Playground opened with the show. It offers early previews of Deep Brush and Hands On, plus Project Custom Filters, which turns a text description into a live Photoshop filter with its own sliders. Adobe lists a conversational character tool and a describe-to-extract sound tool as coming soon. Adobe is asking for feedback at [email protected].
The previews came a day after Adobe shipped its MAX releases, including a Firefly agent that plans the work , Photoshop’s new Relight tool and support for outside AI agents . Adobe has also moved Firefly further into film through its Disney deal .
For Hands On, Adobe says it sees the project as a possible step from text-based editing towards graphical interfaces, and wants to learn whether the problem it solves matters to creatives. The Playground is where it will find out.
Get the most important tech news in your inbox each week.
