TaskbyTask.Practical AI guides for work
ChatGPT · Practical walkthrough

Give your team portraits a consistent style

The team page has twelve photos: three in front of bookshelves, two in a car, one at a wedding. A shared brief for background, light and crop can make them sit together, as long as nobody’s face gets redesigned along the way.

Task by Task12 min read + practice

What you'll make

a short shared style brief for your team page and one edited portrait that the person pictured has approved, ready to pilot on a few more photos.

What you'll need

ChatGPT with image editing (other image tools work, with different controls), one clear original photo, the agreement of the adult pictured, and your organisation’s rules for storing staff photos. Allow about 20 minutes for the first portrait. Each person approves their own result and can reject it.

Example result.

Team directory brief
Background: plain pale mint, soft even light
Crop: square, head and shoulders, full hair visible
Headroom: a little clear space above the hair
Eye height: about two-fifths down, so faces line up in a row
Destination: 240-pixel square cards on the team page

1. Write a shared style brief

Look at the real directory page and decide five things:

  • Background: a plain colour that suits the page and separates hair from backdrop.
  • Crop: how much is visible, such as head and shoulders in a square.
  • Headroom: the space between the top of the hair and the edge.
  • Eye height: where the eyes sit, so neighbouring faces line up.
  • Destination: the real card shape and display size.

Start in plain words and test the brief in the layout before adding percentages or colour codes. Those are targets to check, not promises of pixel-perfect output.

Keep each person’s face, age, clothes and identifying details as photographed. You are arranging a directory, not conducting an audition.

Why this works: consistency comes from the shared settings, not from making people look alike.

What you should see: five short lines you could hand to anyone editing a team photo.

2. Apply the brief to one original photo

Upload the untouched original in ChatGPT and adapt this prompt:

Edit the attached original photo for our team directory.
Preserve this person's facial proportions, apparent age, expression,
hair, clothing, accessories, body proportions and natural skin texture.
Do not beautify, slim, smooth skin or invent features.

Apply our shared style: [background and lighting].
Use [crop and shape], with [headroom and eye height].
The image will appear [destination and size].
Keep the original pose as closely as the crop allows.
Do not add props, text, logos or badges. Make one image.

The bracketed parts come from your brief in step 1. Keep them identical for everyone; only the photo changes. For more on the preservation wording, see writing a headshot brief that says what must stay.

Why this works: the first paragraph protects the person; the second carries the shared style. Keeping them separate stops the style quietly overriding the face.

What you should see: one square portrait with the new background and light, and the same person in it.

3. Check the likeness and the layout

Compare the edit with the original at full size, then ask the person to approve their likeness. Then put it on the real page beside another portrait: are head size, eye height and headroom close enough?

For a crop-only miss, crop or resize the edited image in an ordinary editor. Don’t regenerate a good face to move it a few pixels; each regeneration is another chance to change it.

Why this works: likeness and layout are separate checks, and only the person can do the first one.

What you should see: an approval from the person, and a portrait that sits comfortably in the row.

4. Pilot on a few before the whole team

Before editing everyone, try the same brief on three or four permitted photos that differ the way your team does: glasses, hairstyles, lighting, framing. Place them side by side at directory size and agree acceptable differences, such as similar head size and enough headroom for every hairstyle, rather than identical faces.

If a source is too dark or tightly cropped, ask for a new photo rather than asking the tool to invent what’s missing.

Why this works: one photo can’t show whether the brief works across different people; a small pilot can.

What you should see: a row of pilot portraits that line up, each approved by its own person.

Check the result

  • The person pictured has approved their own likeness.
  • Face, age, hair, clothes and accessories match the original.
  • Background, headroom and eye height match the brief closely enough at real size.
  • Any differences were fixed by cropping where possible, not by regenerating.

The usual failure: the style lands but the framing doesn’t. The background is right, while the head sits higher or lower than the brief asked, so the row looks uneven. Spot it by placing the portrait beside a neighbour at directory size and comparing the space above the hair. If there’s enough image to crop, crop it. If not, go back to the original with one limited change:

Keep the shared background and light. Leave a little more space above
the hair. Preserve this person's appearance, clothing and expression.

Recheck the likeness afterwards, because any regeneration can change a face.

Worked example: the bright directory style

Fictional example. One fictional adult, edited with the bright directory brief.

Unchanged fictional source portrait

Bright directory treatment of the same fictional adult on a pale mint background

The pale-mint background and soft light arrived as asked. The framing didn’t: the eyes and headroom are not where the brief put them, and the image came back at 1254 × 1254 pixels instead of the 1024 requested. Because the face was right, the fix was a crop and resize, not a new generation.

What mattered:

  1. The face survived. Hair, top, earrings and expression match the source, so the likeness check passed first time.
  2. Framing was the miss. It only shows when the portrait sits next to another one at the real size.
  3. Cropping beat regenerating. An ordinary crop fixed the layout without risking the likeness.

The gallery at the top of the page shows the same person in two other styles. A charcoal backdrop kept the navy top visible but the crop differs from the bright version, so matching instructions didn’t make the images line up. The office version invented a plant and a wall, so use real workplace photos if the page implies the location is real.

Try it

Task by Task.Explore the example

Compare the full source and treatment

Move the divider to compare the source and the bright treatment at the same position. Check the headroom and eye height as well as the background.

Unchanged generated fictional portrait sourceActual generated fictional portrait treatmentSourceTreatment

Half of the treatment is revealed over the source. These are generated fictional images, not a real person.

Fictional source and bright directory edit. The background and light changed as asked; the space above the hair did not.

Task by Task · Illustrative example

Task kit for this job: Download (ZIP) · Reusable instructions

Go deeper

Brief a colleague whose photo won’t work

When a source is dark, blurry or cropped through the head, no prompt will recover it reliably. Ask the AI to say what it can’t fix and draft a short capture note for the colleague instead.

This photo is too [problem] to edit for our team directory. List
what an edit could not fix reliably. Then write a friendly, five-line
note asking for a new photo: plain background, facing a window,
camera at eye level, head and shoulders, and no filters.

[problem] is what’s wrong, such as “dark and cropped at the forehead”.

Keep an approval record

For each portrait, keep the original, the prompt, the tool and date, the output file, its real dimensions and the person’s approval. Store photos under your organisation’s rules, with access limited to the people doing the work. Your ChatGPT account’s data controls don’t replace those rules.

A record tells colleagues which image is current and makes the next new starter easier. It also beats a folder of twelve files called final-final.

Sources

Optional analytics

With your permission, Google Analytics uses cookies to measure visits and which guides people read. It stays off until you accept. Use Analytics choices in the footer to change your choice. Rejecting after accepting refreshes this page. Privacy details.