Canva, with ChatGPT, Claude, Copilot or Gemini · Practical walkthrough

Turn one campaign visual into three useful sizes

The event banner looks great in the email. Then someone needs it square for the feed and tall for a story, and the quickest-looking route is to ask AI for "the same image, but 9:16". That's how the speaker gets a new face.

Task by Task7 min read + practice

What you'll make

three exports of the same campaign image (wide, square and tall), each cropped, padded or extended from one master, with the headline as editable text.

What you'll need

the master image you’ve chosen, the sizes each channel needs, a design tool such as Canva, and optionally your work AI account to plan the layouts. About 20 minutes. Canva’s Resize is on paid plans; on the free plan, start a new design at each size and place the master image in it.

Example result.

Size	How it was made	Headline
Wide, 1200 × 627	Slight trim top and bottom	On the wall, left of the speaker
Square, 1080 × 1080	Left of the wall cropped away; speaker kept whole	On the remaining wall
Tall, 1080 × 1920	Image fitted, wall extended above	In the extended wall at the top

1. List the destinations, then plan each layout

Write down each place the image will go and its size. Common sizes are 1200 × 627 for a LinkedIn link preview, 1080 × 1080 for a square feed post and 1080 × 1920 for a story. Platforms change these, so confirm each on the platform’s own help page before you export.

Then describe the master image to the AI and ask for a plan, not new images. Name what must stay whole: a face, a product, a logo.

My master image is [description]. I need it at [sizes].
For each size, tell me whether to crop, fit with padding, or extend the background, and where the headline should go.
Keep [protected parts] whole and unchanged in every version.
Don't create new images. If a size can't work without losing a protected part, say so.

[description] is what’s in the image and where, such as “a speaker on the right third, empty wall on the left”; [sizes] are your destinations; [protected parts] are the things that must not be cut or changed.

Why this works: most resizing is a layout problem, not a generation problem. Asking for a plan keeps the AI from solving it by redrawing the picture.

What you should see: one line per size saying crop, fit or extend, with a headline position that doesn’t cover the protected parts.

2. Make each size from the master

In Canva, open the design and choose Resize from the menu bar. Pick or type your sizes, then Copy & resize, so the original stays untouched. In each copy, move and scale the image following the plan, and type the headline as text rather than baking it into the image.

A crop fills the frame by cutting part of the image away. A fit keeps the whole image and leaves space around it, which you can fill with a colour band for the headline. Neither creates detail that wasn’t there.

Why this works: every version uses the same pixels, so the speaker looks the same in all three. Editable text means a typo is a ten-second fix, not a new image.

What you should see: three designs with the same person, the same clothes and the headline as text you can click and edit.

3. Extend the background only where you must

If a frame needs more image than you have, such as sky above the speaker for a tall story, extend the background rather than regenerating the scene. In Canva, select the photo, then Edit › Tools › See All › Magic Expand, and choose a size or drag a crop handle past the edge.

Why this works: extending the edges leaves the middle of the image alone, in principle. Checking it is how you know.

What you should see: new background at the edges, and a speaker who looks exactly as they did in the master.

Check the result

Open each export and confirm its pixel size matches the destination.
Compare the person or product in each version with the master: face, hair, clothes, hands, logos.
View each version at the size people will see it, such as on a phone, and check that the headline is readable and covers nothing important.
Read every word in the image letter by letter.

The usual failure: the AI redraws the subject. You ask for “the same banner at 9:16”, and the result is a new scene: the speaker’s face, jacket and lanyard change, and the headline comes back misspelled. Spot it by putting the tall version beside the master and comparing the face and text.

This version redrew the speaker and the headline instead of reusing the master image.
Go back to my plan: tell me how to make this size from the master using only crop, padding or a background-only extension,
and where to put the headline as editable text. Don't generate a new image.

Fictional example. A chamber of commerce needs its Northgate Business Breakfast banner in three sizes for email, the feed and a story.

Master: 1600 × 900. Speaker standing on the right third, plain wall on the left.
Headline: "Northgate Business Breakfast · Thursday 12 November"
Needed: 1200 × 627 email/link, 1080 × 1080 feed, 1080 × 1920 story
Protected: speaker's face, hands and name badge

First attempt (wrong): asked for “this banner as a 9:16 story”, the AI returned a new picture. The speaker’s beard had gone, the badge read “Nortgate”, and the headline said “Bussiness Breakfast”.

Repair: the repair prompt above. The plan: wide trims a little from the top and bottom, with the headline on the wall; square crops away the left of the wall so the speaker stays whole; tall fits the image and extends the wall above it for the headline.

Repaired exports: the three sizes in the example result at the top of this page, made in Canva with Copy & resize and one Magic Expand for the tall version.

The redraw changed three things nobody asked for. Using the master’s pixels changed none.
The square crop kept the right of the image, not the centre. A centre crop would have cut the speaker’s shoulder.
After Magic Expand, the speaker was compared with the master before export. The new wall was fine; the badge still read “Northgate”.

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

Go deeper

Mark the safe zone in the brief

Platforms often cover parts of an image with buttons or profile names, especially on tall stories. Ask the AI or your designer to keep faces and headlines away from the top and bottom edges of a story, then check the export on a phone. Name protected parts in words (“the speaker’s face and badge”), not just percentages.

Keep one master for every version

Make every size from the same master file. Generating each ratio separately can produce a different face, outfit, product or label each time, even with identical prompts. If the master changes, remake all three sizes from it.

Record what went where

Keep a short list: destination, size, how it was made (crop, fit or extend) and who checked it. When someone asks why the story looks different from the banner, you can answer in one line. The export guide shows the same file checks for a portrait.