A social image is decided by three things before any styling: the ratio the platform will crop to, the empty space you leave for the subject, and where the words sit.
Three things to fix first
The official record is specific about what the model holds steady. The launch page says Images 2.5 is “better at editing only what you’ve asked for, while keeping the rest of the details the same”, and that across a longer conversation it “follows specific editing instructions more reliably across multiple edits”. Both matter for a set of posts: the first image sets the visual, and the rest should be edits of it rather than new pictures.
The help centre lists the controls you actually reach for, starting with the aspect ratio picker, and warns of one limit worth knowing up front: “Image generation may take a few minutes, depending on the complexity of your request.”
Nothing official publishes a recommended size for any platform, a quota, or a per-image price. Treat those as unknown rather than estimating them, and check the platform’s own current specification instead.
Set the ratio first
Ratio is not a finishing step. It decides where the platform crops, how much room the subject gets, and how much of the frame the words can occupy, so fix it before the first generation.
A single clean illustration for a post, one centred subject on a calm
muted background, generous empty margins on every side, no text, no
logos, no watermark. Vertical portrait framing, tall and narrow.
Subject: <describe the subject and the mood in one sentence>The product layer already supports this: “You can also generate images in any aspect ratio. Use the aspect ratio picker to choose a different aspect ratio, or include your desired aspect ratio in your prompt.” Once an image exists, the editor offers the same control again, so a frame that is right in every way but the shape can be regenerated at another ratio instead of being rebuilt from the prompt. Ratio choices and their trade-offs are set out in ratios and sizes.
Two examples: a cover and a slide
The two shapes a social set usually needs are a single vertical cover and a run of squarer frames. Build the cover first, then treat each slide as an edit.
One: a vertical cover
Vertical portrait image, one strong centred subject, soft directional
light, calm background, wide empty space in the upper third reserved
for a title. No text, no letters. Keep the subject on a single axis,
centred horizontally.Check the empty band is genuinely empty before you write anything into it, and that the crop keeps the subject whole on a tall screen.
Two: a slide in a set
Using the same image, keep the subject identical in shape, colour and
position. Re-compose it inside a square frame, keeping the same
background, the same light direction and the same palette. Leave the
lower third empty for a caption. Change nothing else.Compare the slide to the cover side by side. If the light direction or the palette moved, the set will read as two unrelated pictures.
Keep a series on brand
A series stays coherent when each frame is an edit of one master rather than a fresh prompt. Copy the prompt, change the subject description and leave the lighting, background and framing lines untouched; the goal is that the images agree about everything the reader is not looking at.
Series editing is where the documented multi-turn behaviour helps most, because earlier changes are more likely to survive later ones. One restriction is worth remembering: the model is not a layout engine, so ask for one change per turn instead of stacking five.
There is a dedicated article on holding a subject steady across images, character consistency, if the series needs a recurring person, mascot or product.
Where the words go
Text belongs in the last pass, and the reliable pattern is to reserve a zone in the composition and fill it afterwards. Write down the position and the size, not the mood — “a large bold title, centred in the upper third, on one line” is actionable, “modern and elegant” is not.
Add only the title "NEW SEASON" in large, bold, white sans-serif
letters, centred in the top third of the image. Keep the title on one
line. Do not add any other text.One official limit applies here, and it decides how much you trust a single pass: “Highlights are not always precise, and edits may extend beyond the area you selected.” Check the boundary of the lettering, not just the letters. The full set of layout rules, including how to describe type and hierarchy, is in words inside the frame.
Platform ratios
Specifications differ between platforms and change, so keep them as a per-platform setting rather than a fixed number.
| Use | Ratio to generate | Before publishing |
|---|---|---|
| Feed square | Square | Subject reads at thumbnail size |
| Vertical cover | Portrait | Title band is empty and inside the crop |
| Wide link preview | Wide | Subject is not cut at the left or right |
Before you publish
Run the same short check on every frame in the set. Is the subject still legible at the size the platform actually shows? Is the reserved band for a title still empty? Does the frame carry any unintended logo, watermark or third-party mark? And do the frames agree on light direction, palette and framing? A set that fails the last check is usually cheaper to fix by editing one frame and re-deriving the others than by regenerating the whole run.
Frequently asked questions
Can I skip the cover?
You can, but starting from one master frame is what keeps a series consistent. Build the strongest image first, then derive the rest from it as edits.
Why does the subject move?
Because a new generation composes a new picture. Ask for the change as an edit, restate that position, shape and colour stay identical, and change one variable per attempt.
Must I put the size in the prompt?
Only if the picker is not available to you, since the prompt accepts a ratio in words. Either way, confirm the result is the ratio you intended before building a set on top of it.
Is there a per-post quota?
No official page publishes one, and none prints a price, so treat limits as unknown. Check the current specification for each platform you publish to.
Reuse prompts across a team
Two official mechanisms make a set repeatable rather than personal. Templates offer a starting point for common formats, though the help centre notes they are “not yet available in Work mode”. And a shared image can carry its original prompt, so a colleague can apply the same recipe to their own subject.
Keep the prompts that worked with the images they produced, and reuse them as a house style: the ratio line, the lighting line and the framing line change rarely, while the subject line changes every time. Prompt sharing covers the mechanics, and sticker packs and grids covers the repeated-tile case when a post needs a grid rather than a single frame.