How We Run AI Pre-Shoots for Clients (The Agency Version)

The mechanics of planning a shoot with AI are the same whether it is yours or a client's. Those are covered in the companion guide, and this one assumes you have read it or already know that part.

What changes when a client is involved is everything around the mechanics. Whose face it is. Who sees the previews and what they think they are looking at. What is in the agreement. What you are actually selling.

That is what this covers.

Where this fits in the process

For us, the AI pre-shoot sits between creative direction and the shoot itself, about a week out. Early enough that the plan can still change, late enough that the direction is settled and we are not generating images of an idea nobody has approved.

Running it earlier than that produces beautiful pictures of the wrong strategy.

The order:

1. Discovery and creative direction, as normal
2. Mine the meeting archive
3. Build the client as a character, with permission already signed
4. Generate the full shoot
5. Internal review, then revise
6. Present to the client
7. Revise again from their notes
8. Deliver the shoot plan, book the day

Step 1: Mine the meeting archive first

This is the part that has nothing to do with images and returns the most.

Point Claude at every meeting note you have with that client and ask what they have already told you about how they want to be seen.

Read every meeting note in this folder. Pull out anything the client has said about how they want to be photographed, how they want to come across, what they disliked about past shoots or past photos of themselves, and any specific clothing, colors, or settings they mentioned. Quote them directly where you can. Flag anything they said more than once.

What comes back is never a summary. It is the sweater they mentioned twice, the fact that they keep asking to look approachable and keep getting shot like a CEO, the thing they hated about the last photographer and mentioned once, in passing, eight months ago.

Half of a shoot plan is already sitting in your meeting notes. Nobody goes back and reads them, which is the entire opportunity.

Two things this buys you beyond the plan itself. The client feels heard, because you are repeating their own words back as decisions. And you stop re-asking questions they have already answered, which is the fastest way to look like you have not been paying attention.

Step 2: Get likeness permission in writing, before anything

You are building a client's face into an AI tool. Handle this properly and it is a non-event. Handle it late and it is the worst conversation of the project.

What they need to know and agree to, in writing:

• Their likeness will be used to create a consistent character in an AI image tool, named
• The purpose is internal planning and previsualization for their shoot
• The generated images are not for publication and will not be delivered as final assets
• Where the character and the images are stored, and for how long
• That they can ask you to delete it, and what happens when they do

Put it in the project agreement or a one-page addendum rather than an email. It costs a sentence at the point where everyone is signing things anyway.

[OPEN QUESTION: is this what you actually do? The section is written as current practice. If the permission conversation happens verbally, or lives somewhere else, tell me and I will describe the real process. This is the paragraph most likely to be quoted back at us.]

A related judgment call worth making once, as a policy: whether you tell the client's photographer that AI previews exist. We do. Some photographers read it as being handed a plan and some read it as being second-guessed, and knowing which one you are working with is useful before the day.

Step 3: Generate against the direction, not against your taste

Working for yourself, you generate what you like. Working for a client, you generate what the direction says, including the parts you would have done differently.

This matters because the previews are about to become an approval artifact. If you generate your preferred version instead of the approved one without saying so, you have moved the goalposts without telling anybody, and the conversation you have next is about your taste rather than their shoot.

Generate the approved direction properly first. If you think there is a better version, generate that too and present it as an option, labeled.

Step 4: Review internally before the client sees anything

Never send a raw generated set. The hit rate is not high enough and a bad frame does more damage than a missing one.

Cull hard. Show the strongest two or three variations per look, not the eleven you made. Anything with mangled hands, invented text, or a face that drifted off the character goes in the bin, because the client will fixate on it and you will spend the call explaining AI artifacts instead of discussing the shoot.

The internal review questions are the same ones from the universal guide, asked on the client's behalf:

• Which look gets cut if the day runs long?
• Is any of this just their last shoot again?
• What is missing that they will need in six months on a sales page?
• Does this actually look like the person, or like a stock model with their haircut?

Step 5: Present previews so nobody mistakes them for the product

This is where agencies get burned, and the fix is a single sentence said out loud before anyone scrolls.

"These are planning images, not photos. They exist so you can change your mind about the looks before we spend the day. The real ones will look better than these."

Say it first. Say it again at the end.

Without that framing, two failure modes show up. Either the client thinks the AI images are the deliverable and wonders what they are paying a photographer for, or they judge the creative direction by the quality of an AI render, which is a very unfair test of the direction.

Present them look by look next to the intent, rather than as a gallery. The question you are asking is "does this look serve you," not "do you like this picture."

And decide in advance how you will respond when a client asks whether they still need the shoot. It happens. Our answer is that the previews are a plan and the photographs are the asset, and that AI hands, AI text, and AI fabric fall apart the moment anyone looks closely at a hero image.

Step 6: Deliver a page, not a deck

We used to send shoot plans as decks. Functional, fine, nothing wrong with them.

Then we paid attention to how clients actually open things, which is on a phone, usually standing up, between two other tasks. A sixteen-slide deck on a phone is a pinch-and-zoom exercise, so the work is good and nobody sees it properly.

Now it is a small web page. Schedule, looks, wardrobe, references, the previews, the shot list, all in one scroll, and it resizes because it is a website. Same information, roughly the same build time on our end.

The photographer gets the same link. So does hair and makeup. One artifact, everybody working from it, nobody hunting for the latest version of a PDF.

What this changes commercially

Three things worth knowing before you decide whether to adopt this.

Approvals get faster and better. A client approving a written creative direction is approving a paragraph. A client approving previews is approving something they can actually see, which means the objections arrive a week early instead of on the day.

The shoot day gets shorter and calmer. The photographer receives a shot list that has already survived a client review. Nobody is workshopping looks while a studio meter runs.

Reshoots drop. This is where the money is. A reshoot costs the studio, the crew, the client's second day, and your reputation.

[OPEN QUESTION: do you want to name what this is worth? The strongest possible version of this section quantifies it: what a shoot day costs, how many reshoots you have avoided, and whether the pre-shoot is billed as a line item, folded into creative direction, or absorbed. Any of those numbers would make this the most persuasive thing on the site for an agency reader. All of them are yours.]

The mistakes worth avoiding

• Generating before the direction is approved. You will fall in love with a look nobody signed off on.
• Sending the raw set. Cull to your best two or three per look.
• Leading with the tool. Clients do not care that it is Magnific and Claude. They care that they get to see the shoot early.
• Skipping the framing sentence. Say "these are planning images, not photos" before they scroll, every time.
• Treating previews as a deliverable. The moment you deliver AI images as assets, you are in a different business with different legal exposure.
• Building a character without written permission. There is no version of this that is worth the shortcut.

The client version of this is the same trick as the personal one, with more care around it. Generate the shoot before the shoot. Show them while changing their mind is still free.

Planning your own instead? The companion guide covers it start to finish.