The Reference Photo Shot List
Marketing

The Reference Photo Shot List

The photos to take before you ask AI for pictures of yourself or your team, so the pictures look like the real people.

๐Ÿš€

Want AI pictures of your team that actually look like them? Get in touch and tell me where you post.

Why one good photo isn't enough

AI image tools can put a real person into a scene you never photographed: you at a desk with a client, your team behind the counter, a trainer in the middle of a session. They do it by copying from the photos you give them. Whatever they can't see in those photos, they guess, and the guess is an average person rather than you. If you're very fit, very tall, tattooed, older than average or known for one particular thing, average is wrong in exactly the way people notice first.

I learned this in November with a personal training client. His weekly Facebook posts carry AI pictures of him training clients, made from real photos of him, and he asked for his co-trainer to appear in some of them too. He sent 12 photos of her. The first test used one where she wore long sleeves, and she came out with bare, clean arms, though she has full tattoo sleeves. The model had never seen her arms, so it drew average ones.

Only one of the 12 photos showed her arms, so the second test used that one. The tattoos came out right, but her face and build drifted, because a single photo left the model guessing about everything else. The third test gave it three photos, each with a named job: one for the tattoos, one for the face and a side view for her build. It passed every automatic check on the first try. It still wasn't quite her, so she stays out of the posts until the right photos come in, and the list of those photos is below.

๐Ÿ’ก

Test it with your own photos first. Seeing yourself drawn slightly off is the fastest way to understand why the people in your pictures care about details you'd wave through.

The shot list

Take these in one session, in daylight, with a phone camera. Photos work better than frames pulled from a video, because a video frame is blurrier and gives the model less to copy. Ask for them by name, since "send me more photos" usually gets you twelve of the same kind, and a list gets you the ones that matter.

1. Four or five face close-ups

The face should fill most of the frame, from the chin to just above the hair. Take one straight on, one turned a little to the left, one a little to the right and one or two with a natural smile. Skip sunglasses, hats and beauty filters, because the model copies those too.

2. Three or four full-length photos, front and side

Stand against a plain wall with the whole body in the frame, once from the front and once from the side. Wear clothes that show your real build, close-fitting rather than baggy, because a baggy shirt tells the model nothing and it fills the gap with average. For a trainer that means a tank top with shorts or leggings, and for most offices it's your usual work clothes, as long as they fit.

3. One clear photo of each thing people know you by

Think of what someone would say when describing you: tattoos, glasses, a beard, a uniform, a favorite cap. Give each one its own clear, close photo. Describing a feature in words doesn't work on its own, as the clean-armed first test showed, because the model draws what the photo shows.

4. Plain clothes you'd really appear in

The model copies clothing faithfully, logos included. Her real tank top carried a sportswear logo, and the pictures copied it even with a line in the prompt saying no logos. Wear plain clothes for the shoot, or clothes with your own business name on them.

5. Even daylight, no filters

Shoot near a window or outside in open shade, where the light is soft and even. Hard shadows and filters change the face the model learns from. Look at each photo at full size before you send it, and drop any that are blurry.

Send this to everyone who'll be in your pictures
The Reference Photo Shot List

[ ] 4 or 5 face close-ups: straight on, turned a little left, turned a little right, one or two smiling. No sunglasses, hat or filter.

[ ] 3 or 4 full-length photos against a plain wall, front and side, in close-fitting clothes.

[ ] 1 clear, close photo of each thing people know you by (tattoos, glasses, beard, uniform).

[ ] Plain clothes you'd really appear in, with no big logos.

[ ] Daylight or open shade, no hard shadows.

[ ] Photos, not frames from a video. Every one sharp at full size.

Give each photo a job

If your image tool lets you attach several photos to one request, attach the whole set and say in the prompt what each one is for. The one I use takes several at once, and naming a job for each turned the choice of one photo into a combination. Write the person's distinguishing features down as plain facts too, so the model has both the picture and the words.

Fill in the brackets, attach the photos
I'm attaching [number] reference photos of [name]. Use them like this:

Photo 1: the face. Match the face exactly.
Photo 2: [the tattoos / the glasses / the beard]. Copy them exactly as shown.
Photos 3 and 4: build and height, from the front and the side. Keep the build exactly as in these photos, never heavier, thinner, older or younger.

Features that must always appear: [for example, full tattoo sleeves on both arms, thin black glasses]
Things that must never appear: [for example, logos, a beard [name] doesn't have]

Now make this picture: [describe the scene].
No readable text anywhere in the picture.

Rules belong to the person

In that same system, the main trainer has no tattoos, and an earlier picture once gave him full sleeves anyway. So his pictures get rejected when tattoos appear, and hers get rejected when they don't. If you put more than one person into AI pictures, keep a short list of facts for each one and check every picture against that person's list, never a general one.

Let the person sign off

The person in the picture is the only one who can say it looks like them. An automatic check can confirm the tattoos are there, but it can't tell that a face is slightly off or that someone looks a few years older than they are. My own LinkedIn pictures come from a folder of 21 photos of me, and I still look at every one before it goes out.

For the person in the pictures, before anything is posted
Before my AI pictures go online

[ ] The face looks like me.

[ ] My build and height look like me.

[ ] Everything people know me by is there (tattoos, glasses, beard).

[ ] Nothing is there that I don't have.

[ ] The clothes are something I'd really wear.

[ ] I'm comfortable with this going online with my name next to it.

If any box stays empty, leave that person out of the pictures until better photos come in. Keep posting with the people whose pictures pass. A week without someone costs nothing, and a picture they dislike of themselves costs trust that's hard to win back.

โœ…

Ask everyone who'll appear in your pictures for permission first, and take their photos the same week, so the set matches how they look now.


Andrew Voskov, founder of Cherry Pi AI

About Andrew Voskov

Andrew Voskov is the founder of Cherry Pi AI. He spent twenty years running online businesses before AI, including product brands sold on Amazon, and now helps small businesses with marketing, bookkeeping and custom AI tools. A weekly posting system with AI pictures of a client's real team is one of the custom AI tools he builds.


๐Ÿš€

If you'd like weekly posts with AI pictures of your real team, get in touch. Tell me who would be in the pictures and where you post.

Found this useful? Follow along on LinkedIn. I post free systems and breakdowns every week.

Follow Andrew on LinkedIn