How to Create Banners and Social Media Posts with AI

Generating a pretty image with AI is easy. Generating a social media piece that actually works is another thing — and the difference lies in something almost no tutorial covers: a social piece isn't an image, it's an image plus text plus a visual identity, competing for attention in a crowded feed.

This guide covers the complete workflow: the strategic decision to have AI generate the base rather than the whole finished piece, 25 ready-made prompts organized by goal, how to keep visual consistency across posts, and the production process that turns a generated image into a publishable post.

If you want the general logic of prompt building first, the guide on how to write AI prompts has the complete structure.

Why most AI-generated posts don't work

Three problems keep repeating, and they all come from the same cause: treating the social piece as if it were a standalone illustration.

There's no room left for the text. Almost every social piece has words on it. If the image is visually busy in all four corners, the text has to be thrown over the subject, and the result comes out illegible or ugly. Pieces that work reserve a clean area from the moment they're generated.

Every post looks like it's from a different brand. Images generated in different sessions have different palettes, styles and textures. In the feed, this destroys recognition — someone scrolls past three of your posts without realizing they're from the same profile.

The image doesn't survive real-world size. On a phone, a post appears just a couple of centimeters wide. Compositions full of fine detail turn into noise. The test is always the same: shrink it and look.

💡 Feed test: put your piece next to five posts from competitors, all at thumbnail size. If yours doesn't stand out or isn't identifiable as yours, the problem is the piece, not the reach.

The decision that comes before the prompt

The most important question in this article: do you want AI to generate the finished piece or the visual base?

Generating the finished piece, with the text baked in, is tempting and works for occasional use. But it creates three problems in ongoing production: you don't control the font size, you can't test title variations without regenerating the whole image, and the typography changes with every generation, breaking the brand identity.

Generating the visual base — an image with no text, with a clean area planned in — and applying the text afterward solves all three. You keep the same font across every post, swap the call-to-action in seconds, and reuse the same base in different formats.

The practical recommendation: use AI for the base on anything recurring, and save text-baked-in generation for one-off pieces where the typography doesn't need to be the brand's own.

The six types of piece and what each one calls for

Type	Visual priority	Where to leave space
Feed post	One single focus, high contrast	Center or lower half
Carousel	Coherence across the screens	Wide uniform area on every screen
Story and Reels	Vertical, legible in motion	Middle third; top and bottom get covered
Cover and banner	Horizontal, with side breathing room	Left or right side
Ad	Immediate clarity of the offer	Half the frame for text and button
Profile cover photo	Survives the platform's crop	Generous margin on every edge

The story format deserves special mention: interfaces cover the top with the profile name and the bottom with interaction buttons. Placing important information in those bands is the most common mistake with this format.

The structure of a social piece prompt

1. Subject or visual base

One single recognizable element, or a background texture. Social pieces don't handle complex scenes.

2. Planned free area

Where the text will go. "Clean uniform left half" or "plain center with elements only at the edges". This is the block that separates an image from a piece.

3. Brand palette

Two or three colors, named. This is what guarantees the posts look like they're from the same house.

4. Fixed visual style

The same phrase across every piece — "flat vector illustration", "lifestyle photography with natural light", "minimalist 3D render". Varying this destroys the identity.

5. Format

Square, vertical 4:5, vertical 9:16, or horizontal. Always stated before generating.

Backgrounds and textures for text posts (1–5)

The most useful category of all, and the most overlooked. These are neutral bases you apply text over — reusable dozens of times.

1. Background with leaf shadows

Abstract background of beige paper texture with a soft diagonal leaf shadow crossing it, warm natural light, square composition with a clean uniform center for text, neutral earthy palette, elegant minimalism, no text at all

2. Gradient with grain

Soft gradient background in [color 1] and [color 2] with a diffuse transition, fine grain texture over the whole surface, no identifiable elements, uniform luminosity at the center, square composition, social media post base

3. Geometric shapes at the edges

Composition with flat geometric shapes in [color 1] and [color 2] concentrated only at the top and bottom edges, completely plain center in [background color], modern graphic style, square composition, no text

4. Fabric texture

Macro texture of raw linen fabric with visible weave and soft folds at the bottom edge, raking side light, plain uniform upper two-thirds, beige and sand palette, vertical composition 4:5, post base

5. Overhead desk

Workspace seen from above with a notebook, pen and cup grouped in the lower half, light wood surface, diffuse natural light, empty uniform upper half, warm neutral tones, square composition, clean lifestyle photography

Product posts (6–9)

6. Product on a colored background

[Product] centered on a vibrant solid [color] background, soft cast shadow, uniform studio lighting, plenty of empty space around the object, square composition, clean e-commerce aesthetic, no text

7. Product in a lifestyle scene

[Product] resting on a light marble surface with natural elements beside it, blurred beige-toned background, diffuse natural light from the left, elements concentrated in the lower half, vertical composition 4:5, lifestyle photography

8. Launch on a dark background

[Product] in close-up on a black gradient background, side rim light defining the contour, controlled shine with no blowout, object in the right half and dark empty left side, square composition, premium aesthetic

9. Combo or kit

Overhead composition with [items] arranged in an organized, balanced way on a [material] surface, generous spacing between the objects, uniform diffuse light, emptier central area, square composition, clean flat lay

Educational and carousel posts (10–13)

Here consistency across the screens matters more than the beauty of each one. Use the exact same background and style description on every one.

10. Carousel cover

Solid [color] background with one large, simple geometric shape in the lower-right corner in [accent color], rest of the frame completely plain, flat graphic style, high contrast, vertical composition 4:5, no text

11. Carousel inner screen

Solid [color] background with a thin [accent color] strip at the base, simple flat icon in the upper left, rest of the frame plain and uniform, same graphic style as the cover, vertical composition 4:5, no text

12. Conceptual illustration

Flat minimalist vector illustration representing [concept], simple shapes and few lines, three-color palette in [colors], uniform [color] background, centered element with plenty of negative space, square composition

13. Visual comparison

Composition split vertically down the middle, left side in cool faded tones, right side in warm vibrant tones, same scene of [subject] on both sides, sharp dividing line at the center, high contrast, square composition

Banners and covers (14–17)

Horizontal format calls for a side composition: element on one side, breathing room on the other.

14. Website banner

Wide panoramic horizontal banner with [element] occupying the right third, gradient background from [color 1] to [color 2], clean uniform left two-thirds for text, soft lighting, modern clean style, no text in the image

15. Profile cover

Wide horizontal composition with a soft abstract texture in [colors], no identifiable elements, lighter uniform central area, generous margin on every edge to survive cropping, minimalist style, no text

16. Promotion banner

Horizontal banner with diagonal geometric shapes in [color 1] and [color 2] over a white background, soft shadows, clean central area reserved for text, modern flat graphic style, high contrast, wide aspect ratio

17. Event cover

Horizontal banner with a blurred photograph of [setting] in the background in dark tones, dark gradient overlay on the left half for text legibility, warm light in the background, inviting atmosphere, no text in the image

Stories and Reels (18–21)

Vertical, with the middle third protected. Top and bottom get covered by the interface.

18. Behind-the-scenes story

Vertical composition of [scene] with elements concentrated in the lower half, clean uniform upper third, diffuse natural light, warm neutral tones, spontaneous behind-the-scenes aesthetic, 9:16 composition, no text

19. Quote story

Vertical background with artisanal paper texture in [color], subtle organic shadow on one of the edges, completely plain uniform center, neutral palette, minimalist editorial style, 9:16 composition, no text

20. Product story

[Product] positioned in the lower half on a [material] surface, [color] gradient background, empty uniform upper half, soft studio light, vertical composition 9:16, clean e-commerce aesthetic

21. Reels cover

Vertical composition with [element] in close-up occupying the lower half, dark gradient background above, high contrast between element and background, middle third with space for large text, saturated colors, 9:16 composition

Ads (22–25)

Clarity above aesthetics. The viewer needs to understand the offer in under a second.

22. Product ad

[Product] in close-up occupying the right half on a vibrant solid [color] background, defined shadow, strong uniform frontal lighting, completely plain left half, high contrast, square composition, no text

23. Service ad

Lifestyle scene of [situation] with warm natural light, people in a medium shot on the right slightly blurred, plain uniform wall on the left, warm neutral tones, square composition, authentic photography

24. Discount ad

Background with large flat geometric shapes in [color 1] and [color 2] at the edges, completely plain white center, bold graphic style, high contrast, sense of urgency with no visual clutter, square composition, no text

25. Before-and-after ad

Square composition split vertically down the middle, same scene of [subject] on both sides with different lighting and treatment, sharp white divider at the center, high contrast between the sides, no text

Visual consistency: the brand kit

This is the part that turns loose images into a brand presence. Build a fixed block of text and paste it at the end of every prompt, word for word:

palette in [color 1], [color 2] and [neutral color], style [flat vector illustration / lifestyle photography with natural light / minimalist 3D render], soft uniform lighting, clean composition with plenty of negative space

Three elements need to stay identical across posts: the palette, the visual style, and the light treatment. Only the subject changes. Rewriting this block with synonyms — "navy blue" becoming "dark blue" — is already enough for the result to start drifting.

If you don't have a defined palette yet, generate the piece you like best and extract its colors: the hex codes become your reference for everything else. To choose combinations that work, the guide on color psychology in marketing helps you decide with judgment.

Where the text really comes in

As mentioned at the start, the recommendation is to generate the base with no text and apply the words afterward. In practice, that means adding text over the image, controlling font, size, color and position.

Three legibility rules that apply to any piece:

Contrast above everything. Light text over a dark area, or the reverse. Mid-tone text over a mid-tone background is illegible at thumbnail size.
Few words. A five-to-eight-word call-to-action works better than a full sentence.
Clear hierarchy. One large element, one medium, nothing else. Three different sizes is already too many.

When the background doesn't allow enough contrast, the solution is a solid color strip or a partial darkening behind the text — which can also be requested at generation time.

Formats and sizes

Always generate at the final aspect ratio. Cropping afterward destroys the composition and wastes resolution, especially in vertical formats.

The aspect ratios that cover almost everything are 1:1 for a square feed post, 4:5 for the vertical format that takes up the most feed space, 9:16 for stories and Reels, and a wide horizontal for covers and banners. The exact pixel dimensions vary by platform and change frequently — the guide on the right photo size for Instagram, Facebook, TikTok and YouTube has the updated numbers, and for specific feed adjustments there's the walkthrough on how to crop an image for Instagram.

Negative prompt for social pieces

text, letters, words, watermark, signature, logo, frame border, edge, cluttered composition, elements in the center, small details, too many objects, low contrast, visual noise

The most valuable item is "elements in the center" for story pieces, and "cluttered composition" in any format. These are the two flaws that most often keep the text from having anywhere to sit.

The mistakes that hurt the most

Generating the piece with text baked in for recurring production. You lose control over the font, the size, and the ability to test variations.
Not reserving a clean area. With no planned space, the text ends up over the subject.
Varying the style between posts. Destroys brand recognition in the feed.
Ignoring the interface bands on stories. Top and bottom end up covered.
Detail that's too fine. Disappears at thumbnail size and turns into noise.
Generating square and cropping to vertical. Loses half the composition.
Publishing without compressing. Heavy files load slowly and some platforms recompress with more loss.

The complete production workflow

1. Define the kit. Palette, style and light treatment written in a fixed block. Done once, used forever.

2. Generate the base. At the final aspect ratio, with no text, with a clean area planned in. Generate in batches of four and choose.

3. Adjust the framing. If the aspect ratio didn't come out exact, cropping preserves more quality than stretching, and the resizer gets it to the final size.

4. Apply the text. With add text to image, always using the same brand font.

5. Isolate elements, if needed. To paste a product or person over another base, removing the background gives you a transparent PNG.

6. Finalize the file. Convert with the converter and compress with the compressor — always last, after resizing.

💡 Reuse the base. One good neutral base yields dozens of posts with different text. Build a folder with five or six backgrounds from your palette and you'll stop generating a new image for every post.

Checklist before publishing

Was the piece generated at the final aspect ratio?
Is there enough clean area for the text?
Are the palette and style the same as your recent posts?
Does the text have enough contrast with the background?
Can you understand the piece at thumbnail size, in under a second?
On stories, are the upper third and the base free of important information?
Was the file compressed before uploading?

Apply the text to your base

Generate the image with any prompt from this list and add the call-to-action with the right font and size, right in your browser.

Add text to image

Frequently asked questions

Should I ask for the text directly in the prompt or add it afterward?

For recurring production, add it afterward. Current models write words well, but you lose three things by baking the text into the generation: control over font size, which needs to be much bigger than what the AI usually chooses; the ability to test different call-to-action variations without regenerating the image; and typographic consistency across posts, which is what makes viewers recognize the brand in the feed. For one-off pieces where the font doesn't need to be the brand's own, generating with the text baked in works fine.

How do I make my posts look like they're from the same brand?

Build a fixed block with palette, visual style and light treatment, and paste the exact same words at the end of every prompt. Only the subject should change. Rewriting with synonyms is already enough for the result to drift, so copy and paste instead of retyping. Add the same font and the same text position across every piece, and recognition shows up within a few posts.

What aspect ratio should I use on each network?

Four aspect ratios cover almost everything: 1:1 for a square feed post, 4:5 for the vertical format that takes up the most feed space, 9:16 for stories and Reels, and a wide horizontal for covers and banners. What matters is generating directly in the final format instead of cropping afterward, because cropping cuts the composition in bad spots and wastes resolution. The exact pixel dimensions vary by platform and change frequently.

Why does my post look good on the computer but bad on mobile?

Because in the mobile feed the piece appears at a fraction of the size. Fine details, subtle textures and medium-sized text simply disappear. The rule is to limit the piece to one focal point, use high contrast between the element and the background, and always test by shrinking the image to thumbnail size before publishing. If you can't identify the subject in under a second, the piece needs to be simplified.

Does using an AI image hurt reach?

There's no penalty for using an AI-generated image. What platforms measure is performance: attention time, engagement and retention. Two things are worth noting, though. Some networks label synthetic content when they detect generation metadata, which is informational, not punitive. And images that promise something the content doesn't deliver generate quick abandonment, which does hurt reach — regardless of how the image was produced.

Do I need to compress before publishing?

Yes, and do it as the last step of the workflow. Social media platforms recompress everything they receive, and the heavier the original file, the more aggressive that recompression tends to be — which creates visible artifacts. Uploading an image that's already optimized, at the right dimensions and with a moderate file size, usually results in better final quality than uploading the raw file at maximum resolution.