Adobe's Live Set Edits Video in Real Time — and Splits Brand Trust

By Olivia Hart

Adobe showed a research preview on November 11 that lets creators change a subject, transform an environment, or redirect a camera while a scene keeps playing. The tool, Project Live Set, debuted at Adobe's MAX Sneaks session hosted by Terry Crews, and it runs four simultaneous prompts on one clip with no rendering wait.

The demonstration used a black cat. One prompt added a chef's hat. Others swapped the cat for a Shiba Inu, an alligator in a Miami shirt, and a puppy in a tuxedo. Four versions of the same shot appeared side by side, generated in the moment.

Gautham Mysore, who leads Adobe's video and audio research group, framed the problem plainly. "Today, a large part of the video creation process is waiting," he said. "This breaks creative flow."

What is Adobe not saying?

Live Set is an early research preview. Adobe has not announced a ship date or disclosed the underlying models. Sneaks are previews, and some never reach a product — a caveat the company itself acknowledges by design.

Still, the direction is clear. The author of the source piece, director Bryan Elliott of Goodbrain, reports using Adobe's Generative Object Removal inside the Premiere Beta to erase a parked car from a shot the same day. Video is becoming something you shape with words while you watch.

Why critics fear brand trust collapses

The objection arrived fast: for decades, cameras captured reality, and editing only rearranged the pieces. Now a customer testimonial, a product demo, or a CEO address can bend to a prompt in real time. If viewers stop believing what they see, brand storytelling stops working.

Elliott calls that premise wrong. He argues cameras never captured reality. Crews cast people, dressed locations, lit matte surfaces, chose the best of 20 takes, built burgers with toothpicks and a blowtorch, and let the colorist make the sky bluer than it was. Nobody felt deceived, because audiences know a commercial is a performance.

The real question, he writes, is what promise a brand makes — and whether the promise holds.

Where does the law land?

The Federal Trade Commission drew the line in October 2024, when its rule on reviews and testimonials took effect. The rule bans fake testimonials, including AI-generated ones and any that misrepresent a real person's experience.

The FTC's own guidance sharpens the distinction. There is no blanket ban on AI avatars in marketing. An avatar breaks the rule only when the testimony behind it is fake. Synthetic is allowed; fabricated is not.

Elliott sorts every asset into two buckets. Story — a puppy in a tuxedo selling dog food — sells a feeling, and Live Set belongs there. Testimony — a real customer, a battery demo, a CEO apology — sells a claim. Change the claim, and the deal is broken.

In his words: fantasy in a story spot is entertainment; fantasy in a testimonial is fraud.

What do consumers actually want?

Adobe's own consumer research, a global survey of 4,000 customers run with Oxford Economics, shows the split. 46 percent said they do not care whether a brand uses AI as long as their needs are met. One-third said they would stop interacting with a brand upon learning its content was AI-generated.

Both groups, Elliott argues, want the same underlying thing: that the brand delivers what it promised. The second group simply counts human work as part of the promise.

A separate 2024 Adobe survey of U.S. adults found 93 percent said knowing how online content was made or changed matters to them. Adobe's Content Credentials provide the technical plumbing for a provenance record — a "made by humans" label that Elliott predicts will work like "handmade" on furniture or "organic" on produce.

What changes on set?

The economics matter. A client's "one more in the blue jacket" before losing the light historically meant a reshoot: a second day of crew, gear, location fees, and catering. Adobe says film teams told its researchers they want results during a shoot, while they can still adjust the take.

Elliott's five recommendations for brands:

• Write an AI-use policy before the tool ships; teams will have something like Live Set within a year.
• Label every asset as story or testimony before editing begins.
• Lock testimony: no generative changes to what a real person said or a product did.
• Spend reshoot savings on more ideas and more versions tested in the room.
• Treat human-made footage as a premium asset and prove it.

The tool does not replace judgment. Somebody still decides which of the four cats sells the product. In short: trust lives in the promise, not the pixels — keep the promise, and the audience will let you put the cat in a tuxedo.