Adobe sneaks 11 wild generative AI experiments

Editing moving video used to be a nightmare of frame-by-frame rotoscoping. If lighting shifted or a subject turned around, you basically had to start over.

At this year's Adobe MAX Sneaks, hosted by actor Adam DeVine, Adobe research scientists showed off a wave of experimental tools that bring generative AI into video, 3D, typography, and even wearable hardware.

From single-frame video cleanup to an interactive dress

The standout demo was Project Fast Fill. Powered by Firefly, it brings the magic of Generative Fill into video editing. Editors can edit a texture or remove an unwanted object on a single video frame using a text prompt, and the AI automatically propagates that change across the rest of the clip, adjusting for complex lighting and moving surfaces.

Then came Project Primrose, which brought the room to a standstill. Researcher Christine Dierk walked out wearing an interactive dress crafted from flexible, non-emissive digital textiles. With the click of a button, the entire dress shifted patterns instantly, displaying vector designs created in Adobe Illustrator and Firefly.

Audio and video scaling also received heavy upgrades. Project Dub Dub Dub can auto-dub any audio or video into over 70 languages and 140 dialects, preserving the original speaker's exact tone, vocal cadence, and room acoustics. Project Res Up uses diffusion models to upscale grainy, low-res or cropped video clips while maintaining clean temporal consistency.

Tearing down the walls around vectors and 3D

Typography is notoriously slow work. With Project Glyph Ease, a designer only needs to draw three reference letters, and the AI automatically builds the rest of the alphabet into fully editable, scalable vector glyphs.

Adobe also tackled vector generation head-on. While Project Draw & Delight turns rough sketches into polished vector graphics, Illustrator is gaining the official Firefly Vector Model. Trained on tens of millions of vector images from Adobe Stock, it outputs layered, editable objects with clean anchor points rather than messy paths.

Other prototypes rounded out the lineup: Project Neo lets 2D artists handle 3D perspective without touching complex 3D software; Project Scene Change composites footage from different camera angles with matching motion and realistic shadows; and Project Stardust brings object-aware intelligence to photo editing, making individual image elements as easy to grab and move as cutouts.

Why it matters

We think Adobe is playing a very smart hand here. While many AI startups are obsessed with generating complete images or finished video clips from text prompts alone, Adobe is focusing on real-world editability.

For freelance creators and production houses, client work always demands tiny revisions. You rarely need a completely random new scene; you need to change a logo on a moving shirt, match a brand font, or tweak an individual layer. By tying generative models directly to vector anchor points, audio acoustics, and timeline tracks, Adobe makes AI fit the messy realities of commercial production.

There is still a catch, of course. Most Sneaks projects are research prototypes that may take months or years to reach stable Creative Cloud releases, if they ship at all. But by showing how generative AI solves precise micro-bottlenecks rather than trying to replace the artist outright, Adobe looks well-positioned to keep professional creators firmly inside its ecosystem.