Adobe’s Live Set preview changes video during playback

On November 11, Adobe introduced Project Live Set at MAX Sneaks, letting editors prompt changes to a subject, environment or camera angle while footage continues playing. The early research preview has no announced ship date, and some Sneaks demos never become products.

What does the preview show?

Terry Crews hosted the session. Adobe showed a black cat staying in frame while four prompts ran simultaneously in one window: add a chef’s hat, replace the cat with a Shiba Inu, replace it with an alligator in a Miami shirt, and replace it with a puppy in a tuxedo.

The four versions appeared side by side without a wait. Gautham Mysore, who leads Adobe’s video and audio research group, said: “Today, a large part of the video creation process is waiting. This breaks creative flow.”

In a separate test, Bryan Elliott used Adobe’s Generative Object Removal feature inside Premiere Beta, with help from Claude Fable 5.2, to remove a parked car. It showed editing with words, not a Live Set trial.

What could change on set?

Elliott, who runs Goodbrain and directs commercials, documentaries and branded content, cites Dune: Part Two as a volume-production example and has worked on a 30-by-90-foot LED wall. He says daylight-dependent reshoots often consume more budget than other line items, with each attempt returning crew, gear, location fees and catering.

Adobe reports film teams want results during a shoot, while adjustments remain possible. Live Set could show clients alternatives in the room, rather than two weeks later, after talent left and the location was booked. No benchmark for time, quality or money saved accompanies the preview.

Elliott expects story work to become faster, cheaper and stranger, with reshoot savings shifting into ideas, talent and additional versions. Documentary teams could use prompts for title sequences, reenactments and establishing shots at unaffordable locations, replacing a plane ticket with a generation step.

Interviews should remain untouched because they carry testimony. Directors and brand strategists would still select the winning concept, making taste and judgment the scarce inputs and giving trained eyes more leverage.

Where does trust break?

Critics say real-time edits could make testimonials, demonstrations and executive addresses unreliable. Elliott agrees but draws a sharper line. Commercials already stage reality: crews cast people, dress locations, light products, run a line 20 times and color skies bluer. Viewers buy them as performances, he argues.

His framework separates story from testimony. Story assets include a puppy selling dog food, a sneaker on a software-built rooftop, or actors playing a founder’s childhood; viewers buy emotion. Testimony assets include a customer claiming a fix, a battery-life demonstration, or a CEO apologizing for a data breach; viewers buy a claim.

The Federal Trade Commission’s review and testimonial rule took effect in October 2024. It prohibits fake testimonials, including AI-generated ones and portrayals that misrepresent a real person’s experience, creating legal exposure rather than only a trust problem.

FTC guidance does not ban AI-generated avatars outright; messages fall within the rule when underlying testimony is fake or false. Synthetic performance can be permissible; fabricated experience is not.

What do the audience data show?

Adobe and Oxford Economics surveyed 4,000 customers globally. Forty-six percent said they did not care whether a brand uses AI if their needs were met. One-third said they would stop interacting with a brand after finding its content was AI-generated rather than human-made.

Elliott’s refrigerator analogy sharpens the split: buyers care whether an automated grocery list is accurate and untampered, not whether a machine wrote it. Both groups expect the brand to deliver; one also treats human work as part of the promise.

A 2024 Adobe survey of U.S. adults found 93 percent cared how online content was made or changed. That survey’s sample size is undisclosed. Elliott expects a “made by humans” premium label as synthetic video becomes cheap, but that is an author forecast, not measured demand.

Content Credentials can record whether material was altered, giving brands a provenance claim to sell. Elliott also expects off-grid experiences to gain value as screens and AI multiply.

What should brands decide now?

• Set generative-video rules before shipping; Elliott expects comparable tools within a year, but Adobe has not committed.
• Classify assets as story or testimony in every brief.
• Lock testimony against changes to words, product behavior or executive promises; allow sound and color cleanup.
• Use reshoot savings for concepts, talent and more client-side tests, killing weak concepts early.
• Treat real people and locations as premium assets, with provenance proving their use.

“The camera was never the source of trust. Your promise is,” Elliott said. As real-time edits become routine, brands must prove that every altered frame still delivers the promise its audience bought.