The surface
Texture, colour field, atmosphere, an illustrative motif the thing sitting behind everything else. It carries no information, which is exactly why approximation is safe here. Backgrounds that land first time.
Backgrounds in seconds · type stays yours
Muse Image · graphics for feeds
It makes the layer under the words.
Almost every social graphic is a picture with type on it, and type is the one thing Muse Image cannot be trusted with. Accept that and the workflow gets simple: generate a surface with the space already reserved, then set the words yourself in anything that does type properly.
What you are actually making
Three layers, one of them generated
Thinking of a social graphic as a single image is what leads people to ask for the impossible. Split it into layers and it becomes obvious which part the tool is for.
The words
Headline, date, price, handle, call to action. Rendered text is unreliable at every tier and every prompt length, so the words go on afterwards in anything that sets type a design app, a slide, even Preview.
The brand marks
Your logo, your wordmark, your icon set. A generated approximation of your own mark is worse than useless: subtly wrong, everywhere, forever. These are fixed assets you place.
The corollary is liberating: since the tool never handles anything that has to be correct, a merely pleasing result is a finished result which is why social graphics are the fastest thing to make this way.
The core technique
Reserve the space in the prompt, not afterwards
A beautiful image with your headline sitting awkwardly over a busy patch is the standard failure, and it happens because the space was an afterthought. The composition clause of a four-part prompt is where you fix it, before anything is generated.
“A clear empty band across the middle.” “Subject pushed to the right, the left third left plain.” “Detail only at the edges, centre uncluttered.” These phrases do more for a social graphic than any style word.
Decide where the words go first. Then describe the picture as the thing that surrounds them. It inverts the usual order and it is the whole trick.
Say the shape in words. Square, tall and vertical, wide banner. Meta has not published which output sizes or aspect ratios are available, so rather than assuming one, describe the shape, see what you get, and crop deliberately checking what your own app produces is faster than any claim on a page like this.
Then leave a margin past the zone. Feeds crop, platforms change, and a design that only works at one exact ratio will eventually be cut somewhere unhelpful.
Five rules
Making something that survives being small
Social graphics are designed at full size and judged at thumbnail size in a moving feed. Almost every rule below is about that gap.
One idea per image
A single subject, large in frame. Two competing elements become an unreadable smudge at feed size, and the prompt clause that gets you there is a composition instruction rather than a longer description.
High contrast, or nothing
Subtle tonal work that looks sophisticated at full size disappears on a phone in daylight. Ask for strong separation between the subject and the ground, and pick a palette that survives being small and dim.
No fine detail worth looking at
Intricate texture is invisible at the size that matters and is also where approximation errors hide. Simple shapes are both more legible and less likely to embarrass you.
Keep the important part away from the edges
Platforms crop differently and change how they crop. Anything near a border is one interface update away from being cut in half, so the reserved zone belongs comfortably inside the frame.
Check it in context, not on a white page
Drop it next to your last few posts before publishing. A graphic that looks striking alone can be the one that breaks the run which is the grid-level problem.
Six recipes
Post types, with the space already reserved
Each prompt follows the four-clause form, with the composition clause doing the work of holding a zone open. The diagrams show where the type goes they are drawings, not generated output.
Quote card
A soft abstract field in deep green and cream, no text anywhere, grainy print texture, square, with a clear empty band across the middle, flat even light.
Then: set the quotation and the attribution in the band yourself.
Event announcement
A decorative border motif in ink and gold, centre left plain, letterpress illustration, vertical, pattern at the top and bottom only, even light with a slight paper sheen.
Then: add the what, where and when in the clear middle never in the prompt.
Carousel cover
A single bold symbolic object on a saturated ground, no lettering, graphic poster illustration, limited palette, square, object high in frame, lower quarter kept plain, hard directional light for a strong shadow.
Then: title in the lower quarter. Generate each slide separately against one anchor never as a multi-panel image.
Profile or channel header
A simple seasonal motif with nothing written on it, flat illustration, three colours, wide banner, subject to the right, left third left empty, flat light, no shadow.
Then: name and tagline in the left third, and remember headers crop hard on phones.
Product post
A plain stone plinth against a warm neutral backdrop, nothing on it, editorial product photography, square, plinth in the lower third, space above, one soft light from the left.
Then: composite your real product photograph in. Do not generate the product why that matters.
Article or link preview
An abstract visual metaphor for the article’s idea, no words, minimal editorial illustration, two colours, wide, motif in the upper two thirds, lower strip plain, flat even light.
Then: headline in the lower strip, or leave it many platforms add their own title card.
All six are variations on one move: name the zone in the composition clause. Swap the subject and style clauses freely; the zone phrasing is the reusable part. More starting points in the prompt library.
The grid, not the post
Your feed is the thing people judge
Individual posts are seen once, in passing. The grid is seen every time someone decides whether to follow you, and one post that does not belong is visible instantly.
01
02
03
04
odd one
06
Five images made the same way, and one made on a different day with a different prompt. You spotted it before reading this sentence.
Two rules keep a feed together
Fix the wording. Keep the style and light clauses identical across every post and change only the subject. Those two clauses are your visual identity in text form, so paste them rather than retyping them.
Anchor to one image. Use your best result as the reference for everything that follows, and reference that same original file each time rather than chaining off the previous post otherwise the run drifts a little further every week.
Before you post it
Where generated imagery stops being decoration
Illustration is fine. The line is whether a viewer would read the image as a record of something real.
| Post | Where it stands |
|---|---|
| A decorative background behind a quote | Fine. Nobody reads it as evidence of anything. |
| An abstract illustration for a blog post | Fine. Clearly illustrative by nature. |
| A “photo” of your premises, team or event | No. A viewer takes that as documentary, and it is not. Photograph the real thing. |
| A generated image of the product they will receive | No. Shoppers treat product images as a promise use the real photograph. |
| Invented testimonials, reviews or customer faces | No, under any framing. Fabricated endorsement is fabricated whether the face was generated or stolen. |
| A stylised illustration of a real place, clearly stylised | Usually fine, provided the treatment makes it obvious it is a drawing rather than a photograph. |
On labelling: platform rules differ, change often, and some platforms apply automatic labels of their own. Check the current policy for each place you post rather than assuming, and disclose wherever there is any chance a viewer would take the image for a photograph. What Meta says it uses, and commercial use is governed by Meta’s terms rather than by this page.
FAQ
Social graphics questions
Can Muse Image put text on my social graphic?
Not reliably, and this is the single most important thing to know before starting. Rendered text is a known weak point across image generators, so the working method is to generate the picture with a deliberately empty area and set the words yourself in any editor.
How do I leave room for my headline?
Ask for it in the composition clause. Phrases like a clear empty band across the middle, or subject pushed right with the left third left plain, reserve the space before you need it which is far easier than trying to clear it afterwards. The technique in full.
What size or aspect ratio should I ask for?
Describe the shape in words square, vertical, wide banner rather than assuming any particular ratio is supported, because Meta has not published which output sizes are available. Check what your own app produces and crop deliberately afterwards.
Why does my graphic look good big and terrible in the feed?
Because social images are judged at thumbnail size while you designed them at full size. One idea, a large subject and high contrast survive shrinking; fine detail, small elements and low-contrast palettes do not the five rules.
How do I make a whole feed look consistent?
Keep the style and light wording identical across posts and change only the subject, then use your best result as a reference image for the rest. Consistency at the grid level matters more than any single post.
Can I generate my logo?
No treat your logo as a fixed asset you place yourself, never something to regenerate. A generated approximation of your own mark is worse than useless because it is subtly wrong everywhere it appears.
Is it alright to use generated images for a business account?
For decorative and illustrative posts, generally yes, subject to Meta’s terms on commercial use. What to avoid is anything a viewer would read as documentary a photograph of your actual premises, staff, or the item they will receive. Where the line sits.
Do I need to label generated images as AI?
Platform rules differ and change, and some platforms apply automatic labels of their own. Check the current policy for each platform you post to rather than assuming, and disclose where there is any chance a viewer would take the image as a photograph.
What about generating a carousel of several slides?
Generate each slide separately against one anchor reference rather than asking for a multi-panel image in a single request. Multi-panel layouts drift and the panels will not share a consistent look.
Does this cost anything?
No. Image generation is free with rate limits most people never reach, so producing a month of post backgrounds in one sitting is realistic that burst pattern is the one case where heavy image users do get close to a ceiling. Pricing and limits.
Reserve the zone · set the type · check it small
Decide where the words go before you generate anything.
Then describe the picture as the thing surrounding them, shrink it to thumbnail size to check it works, and keep the style wording identical across the whole run.
Independent guide not affiliated with Meta. Platform labelling rules and Meta’s commercial terms both change; check Meta’s own documentation and each platform’s current policy before posting.