Continuity is the new design skill

Generate the same character twice with most image tools and you get two different people wearing the same description. The hair is close. The face isn't. A brand mascot that looks slightly different in every post isn't a brand mascot, it's a coincidence with a name.

The feature that sounds boring until you need it

Adobe's Firefly agent, announced April 15, 2026 in partnership with Anthropic, started as a way to describe an outcome in conversation and have it executed across Photoshop, Premiere, Lightroom and Illustrator without switching apps. That's the headline capability. The one that actually matters for consistent brand work came two months later: the ability to save a specific character, location or object and reuse it, the same one, across separate generations.

David Wadhwani, Adobe's president of Creativity and Productivity, framed the shift as one where "your perspective, voice and taste become the most powerful creative instruments of all." That line does a lot of work. It's true only if something holds the output constant long enough for a perspective to be recognisable in it at all.

Why sameness is suddenly the hard part

An agent generating from scratch every time has no memory of what it made last time, so nothing forces the tenth output to resemble the first. That was fine when each generation was a standalone image. It stops being fine the moment the same character needs to show up in a launch video, a social post and a product photo and be recognisably the same character in all three.

Notice what that does to the order of operations. A brand guideline used to be a document describing how to make the thing correctly. Now it has to be an artefact the tool can hold, because the tool is doing the making. A rule written for a person to read and a reference saved for a generator to reuse are not the same object, and only one of them survives contact with an agent that regenerates from scratch every time.

What this changes about the designer's job

The move that matters most here is deciding, once, what the character actually is, then making that decision reusable instead of re-litigating it on every prompt. Generating the image was never the hard part. That's a shift from producing images to producing a reference the tool can hold onto, which is closer to art direction than to prompting.

It also means the actual bottleneck moves earlier in the process. The hard part stops being "generate a good image" and becomes "decide, correctly, the one time it counts, what this character or asset actually is," because every later generation inherits whatever that first decision got right or wrong.

The five-minute habit

Before your next AI-generated asset that needs to recur, save it as a named reference the moment you're happy with it, rather than regenerating a fresh description each time you need it. Whichever tool you're on, that's a five-minute step now instead of a full design-system document, and it's the difference between a character your audience recognises and a series of close approximations of one.