Adobe Max: Four AI Projects, from 3D Photography to Characters

Adobe showcased several generative AI "sneaks" at Max: a dialogue system for shaping characters, photos converted into 3D scenes, text-guided sound mixing, and identical replicated retouching. These prototypes have no guarantee of release but are part of a strategy where AI plays a central role.

"Sneaks" Without Guarantees but at the Heart of Adobe's AI Strategy

The projects presented at the Max conference are technological demonstrations, and there is no assurance that they will become available features. They could eventually be integrated into Creative Cloud applications or the Firefly platform, but no confirmation has been provided. Most rely on Adobe's advancements in generative AI. AI tools now occupy a central place in the company's growth strategy, and it is deemed likely that at least some of these prototypes will be officially offered. A common goal of these innovations is to simplify and make more intuitive complex processes of photo retouching, video production, and audio mixing. Gautham Mysore, a senior researcher at Adobe, notes that making these interactions more intuitive also aims to provide experienced creatives with more precise control.

Shaping a Character Through Dialogue

Project Screen Test is designed to facilitate the creation and use of AI-generated characters in video content. Rather than being limited to a textual description, the user can engage in dialogue with the character in an audition-like environment. The user starts by assigning personality traits such as "anxious, shy, clumsy" and selects a voice, either from the provided options or by uploading an audio clip. The personality then becomes more defined through exchanges, mentioning the character's backstory, temperament, goals, flaws, fears, or even gestural habits. Screen test animations are produced based on the description of a setting or situation. For now, it is a standalone creative exercise, with no determined integration with Adobe's existing AI video tools.

Entering a Photo and Expanding it in 3D

Project Sight Span transforms a photograph into a navigable and repositionable 3D environment. The system generates objects and backgrounds that extend beyond the initial frame while integrating naturally into the image. It works with any flat image, whether photos, paintings, or illustrations, and retains the original artistic styles. During repositioning, elements like tattoos or facial features remain recognizable and true to the reference.

From Soundtrack to Batch Retouching, Two AI Assistants

Project Sound Bender is an audio mixing tool that generates sound effects from a textual description associated with a video clip and then automatically synchronizes them. A square sound canvas allows for real-time sound adjustments, with each corner corresponding to an AI-generated effect. Project Rinse Repeat, on the image side, copies a modification and automatically applies it to other objects or people in the same photo. It can, for example, replicate retouching of imperfections across multiple subjects or add shadows to products, even if their shapes and sizes differ. Presented at Max, these projects cover uses ranging from transforming images into 3D scenes to creating interactive characters, as well as simplifying retouching operations.