How to Brief an AI Studio
No agency would hand a photographer an unfiltered client brief. Yet that discipline evaporates the moment AI enters the workflow. After a recent campaign we made with an agency partner - both sides sat down afterwards and compared notes honestly. Here's the result of that discussion - what a good AI brief actually looks like, from the studio that has to execute it.
Author: Heath Waugh
TL;DR - AI didn't remove the need for production discipline. It removed the safety nets that used to enforce it. On a traditional shoot, a loose brief gets rescued by stylists, producers and art directors. In AI production, that scaffolding doesn't exist - the brief carries all the weight the crew used to carry. Lock creative direction before production starts, use sign-off gates, treat "amends" with suspicion, and flag complex environments early. The tools are new. The discipline isn't.
——————————————————————————————————
We recently wrapped a campaign with an agency partner, and afterwards we did something that doesn't happen often enough: both sides sat down and wrote out what we'd learned. Not the polished case-study version. The honest version.
It was a long, and at times, difficult job. Everyone was extremely happy with the result, but both sides knew we could have gotten there in a more efficient way. So we talked it through.
Their learnings and ours landed in almost exactly the same place. And nearly all of them trace back to one thing: the brief.
If you're an agency or brand about to brief an AI studio for the first time, this is what we wish every client knew before the job starts.
The brief is the crew now
On a traditional shoot, a loose brief gets rescued. There's a stylist making wardrobe calls, a producer catching logistics, an art director adjusting on set, a photographer who's shot a hundred campaigns filling the gaps. That support system shapes decisions before and during production, and everyone takes it for granted.
In AI production, that scaffolding isn't there. AI production has a different structure. If the brief doesn't specify tone, colour, wardrobe, mood, spatial relationships or art direction, those decisions don't disappear - they fall to the AI artist mid-job, or they get made by omission. Either way, the work drifts somewhere nobody intended.
The more detail captured before the job begins, the smoother and more aligned the process will be. That sounds obvious written down. It's ignored constantly.
Treat AI artistry as production, not exploration
You wouldn't hand a director a raw client brief with no creative direction attached. The same discipline applies here.
Creative direction should be interrogated, shaped and signed off before it reaches the studio. When it isn't, creative exploration happens during production - on production timelines, at production cost. That's the single biggest source of blown budgets and stretched deadlines in AI work, and it's entirely avoidable.
We do our best work executing a clear direction. We can absolutely help develop one - but that's a different phase of the job, and it should be scoped and signed off as one.
"Editable" is a trap
Because nobody is standing on a set, AI work feels infinitely editable. There's no wrap time, no lighting to strike, no talent going home. That perception quietly dissolves the line between pre-production and post-production until the whole job becomes one long soup of changes.
The fix is gates. Creative direction locked before any generation begins. The AI brief locked before prompting starts. A hero key visual approved before the route scales across executions.
Each gate defines what's still open and what's now closed - and changes after final art are new scope, not a favour.
This isn't bureaucracy. It's the same structure every shoot has always had. AI just makes it easier to skip, and more expensive when you do.
"Amends" is the most dangerous word in AI production
Depending on where you are in the process, a requested change might be a superficial edit - or it might be the equivalent of a re-shoot. The word "amends" flattens that distinction into something that sounds like it takes ten minutes.
Here's the technical reality. AI image editing is inherently destructive to the areas around the edit. Unlike Photoshop, where you can layer a new element in cleanly, generative editing reworks the whole surrounding region to integrate the change - which means adjusting one element can shift alignment, colour and texture in areas nobody touched, and degrade quality in that region. Often there are workarounds. Sometimes there aren't, particularly when the background is detailed or carries strong highlights, shadows or texture.
So feedback needs sorting, not just collecting. Is this a regeneration, a retouch, a rollout note, or a nice-to-have? What does it cost in time, quality and approved direction? On the campaign in question, we hit exactly this on complex flooring and seating surfaces - clean isolated edits were genuinely difficult without compromising what was already working.
The practical takeaway: if your scenes involve complex environments, flag it at briefing stage, not feedback stage. Adding extra elements into those types of scenes after the fact is difficult. Including them in the first place is easy, adding them after the fact isn't.
Fewer options, stronger recommendation
A wall of options feels generous, and is a trap we’ve fallen in to before. It isn't. It sets everyone up for decision paralysis and extra rounds. Requests like "Can we take the lady's shoes from this image and the dress from this image" are all too common when you work this way.
The better model: the agency and studio lead with a point of view. One hero route, properly developed, with supporting options used only to explain why the hero is right.
If the creative direction was genuinely signed off, the first round of generation should already be close - and choosing between ten variations is a sign the direction wasn't actually locked.
Know who's in the room
A creative client reads references differently to a marketing client. For a brand guardian, a reference is an example - a rough compass heading. For a client with a creative background, that same reference often becomes an element they expect to see pulled through into the final work.
Neither is wrong. But they're different jobs, and you need to know which one you're doing. Identify who's around the table at kickoff, because it changes what the references mean, what rationale people need, and how feedback should be read.
Show the process before you start it
Everyone has played with consumer AI tools now, and that's created a false intuition: that the output looks like the prompt, and everything happens instantly. The gap between playing around in a consumer app and a professional pipeline — multiple models, individually generated elements, compositing, grading, retouching - is enormous, and telling people that is far less effective than showing them.
One or two real examples from past projects, shown in the first meeting, resets expectations better than any explanation. What a prompt actually produces. Where the limits sit. Why the finished image took the path it did.
The bottom line
None of this is new thinking. It's the same rigour that's governed shoots for decades - locked direction, tight briefs, honest scoping of changes - applied to a medium that makes it tempting to skip.
The agencies and brands that brief AI production like a shoot will get work that holds up on a billboard. The ones that brief it like a chat window will get endless rounds of amends, and a result nobody's proud of.
The tools are new. The discipline isn't.
Heath Waugh
FAQ
What should be included in a brief for an AI studio?
Everything you'd give a photographer, plus the things a crew would normally catch on set: tone, colour, wardrobe, mood, spatial relationships and art direction. If it matters to the final image, it goes in the brief - because in AI production there's no stylist or art director downstream to fill the gaps.
Should creative direction be signed off before AI production starts?
Yes. Creative direction should be locked before any generation begins, the same way you'd never hand a director an unfiltered client brief. When direction gets worked out during production, you pay for creative exploration at production rates - and usually on production deadlines.
Why can't AI images just be endlessly edited?
Because AI editing is destructive to the areas around the edit. Unlike Photoshop layers, generative editing reworks the whole surrounding region to integrate a change, which can shift alignment, colour and texture in areas nobody touched. Some changes are quick edits. Others are the equivalent of a re-shoot - and the word "amends" hides the difference.
How many creative options should an AI studio present?
Fewer than you'd think. One hero route with a clear rationale beats a wall of variations, which mostly produces decision paralysis and extra rounds. If the creative direction was properly signed off, the first round of generation should already be close.
Do complex backgrounds matter in AI production?
More than most people expect. Detailed surfaces, strong shadows and heavy texture make clean isolated edits genuinely difficult without compromising what's already working. Flag complex environments at briefing stage so they can be planned for - not at feedback stage, when it's too late.
Is AI image production faster than a traditional shoot?
Usually, yes - but not instant. Professional AI production involves multiple models, individually generated elements, compositing, grading and retouching. It's a pipeline, not a prompt. Campaigns typically move around 50% faster than traditional production, provided the brief and sign-off gates are respected.
