Photos and visuals is a Skill and Tech job: you write one style brief, turn each post into an image prompt, and an image tool renders the pictures your camera roll cannot cover. The brief and a first batch of eight to ten images take about an afternoon, and every batch after that takes far less. This page covers image prompts and on-brand AI images. For writing the descriptions every image needs, our social media alt text guide covers that part in full.
The AI writes the prompts and renders the images. You decide which posts get an AI image at all, and you check every one before it goes out.
- Shape
- SkillTech
- Sitting
- an afternoon
The shape: Skill and Tech
The Skill half is the writing. A good image comes from a good prompt, and a good prompt comes from a style brief you write once and reuse. That brief names your colors, your light, your setting and the things that never appear. Once it exists, the AI turns any post into a prompt that fits it.
The Tech half is the image tool, which you pick and set up once, then stop thinking about. Some chat tools make images inside the conversation, and some only write the prompt for a separate image tool. Either way works. If your chat tool cannot make images, it writes the prompt and you paste it into one that can, such as ChatGPT.
What this means for you: most of the afternoon goes into the brief, not the pictures. A vague brief gets ten images in ten styles. A specific brief gets ten images that look like they came from the same business, which is the whole point.
What you give the AI
Gather these before you open the chat. Every item is something you already have.
- Your colors, by hex code. Two or three, from your logo file or your website. "Blue" gets any blue. A hex code gets yours.
- Three to five images you already like. Your own best photos, or a screenshot of your website's top section. Upload them if your tool takes images; describe them in a line each if it does not.
- The look, in plain words. Photographic or illustrated, bright or moody, close up or wide, busy or calm. Pick one of each.
- Your setting. Where your work happens: a kitchen, a job site, an office with plants, a backyard, a studio. The AI places scenes there.
- What never appears. People's faces, a storefront, your products, competitors' colors, anything your industry does not allow you to show.
- The posts that need an image. The captions or topics for the next batch, one line each. Our Instagram page writes these from one real customer question.
- The sizes you post at. Square, portrait or vertical, by platform. The social media image sizes guide lists each frame.
If you have no photos you like, use your website's top section. It is the closest thing most small businesses have to a written look.
What comes back
Two things come back, in two steps. First a style brief, a short block of text you save and reuse. Then one image prompt per post, each with the brief built in, plus the images themselves if your tool renders them.
Paste this first, with your material filled in:
Act as the art director for the business described below.
Write a style brief of no more than 120 words that an image tool can follow.
Our colors (hex):
Images we like (attached or described):
The look: photographic or illustrated, bright or moody, close or wide, busy or calm:
Where our work happens:
Never show:
The brief must name the palette, the light, the camera angle or drawing style, the setting and the never-show list.
Use plain, concrete words an image tool can act on.
Do not use words like "stunning", "premium" or "professional", because an image tool cannot draw them.Read the brief, fix anything that is wrong and save it in a note. Then paste the second part into the same chat, with your sizes and posts filled in:
Using the style brief above, write one image prompt for each post below.
For each post:
1. The post's one point, in a sentence.
2. An image prompt that starts with the full style brief, then adds 40 to 70 words that show that point as an object, a setting or an idea.
Leave clear empty space on one side for a headline.
No words, letters or numbers inside the image.
No people's faces, no storefronts, no finished jobs, and nothing on our never-show list.
3. The size, from this list:
4. One line of draft alt text that says what the image shows and why it is in the post.
If a post is better served by a real photo, say so instead of writing a prompt.
The posts:The reply is a numbered list: one point, one prompt, one size and one alt text line per post. If your tool renders images, ask for them one at a time, so you can reject a bad one before the next is drawn. Ask for two or three versions of any image that matters, and pick one.
The line about real photos matters. When the AI says a post needs a real photo, it is usually right, and that post goes on your photo list instead.
How to judge it
You can reject a bad image in a few seconds. Look for these:
- A style that drifts. Image six looks like a different business from image one. The brief was missing from the prompt, or it was too loose. Paste it again and name what changed.
- Words inside the picture. Misspelled signs, warped letters, a fake logo. Ask again with no text in the image, and add your words yourself.
- Hands, faces and fine detail that look wrong. Extra fingers, melted tools, a face that is almost a person. Reject it. A viewer notices before they read the caption.
- An image that makes a claim. A team you do not have, a building that is not yours, a finished job you did not do. Delete it, even if it looks good.
- An image that has nothing to do with the point. A pretty sunrise over a post about invoices. Ask again, and name the object or setting that shows the point.
- Colors near yours, not yours. Hold it next to your logo. If the blue is the wrong blue, put the hex code earlier in the prompt.
- No room for the headline. The subject fills the frame, so any text you add covers it. Ask for the empty space again, on a named side.
- Alt text that describes the prompt. The draft says "a warm kitchen at sunrise" and the image shows a gray counter. Rewrite it from what is on screen.
If most of a batch fails the same way, fix the brief, not the images. One better line in the brief fixes every image after it.
What stays with you
The AI writes the prompts and draws the pictures. These parts stay with you:
- Which posts get an AI image. A real photo of your work beats any rendered image, every time it exists. Our page on not having enough photos covers what to post when the camera roll runs dry.
- Honesty. An AI image never stands in for proof. No invented staff, places, products, jobs or before and after pairs. If an image could pass for a real photo of your business, label it or do not post it.
- The rules on labeling. Some platforms label AI images, and some ask you to. Check each platform's current help page, because the rules change.
- Rights. Each image tool has its own terms on commercial use. Read them once for the tool you use. Never prompt with another business's logo, a real person's likeness or a named artist's style.
- The words on the image. Add headlines and text yourself, in a design tool where you can check every letter.
- The alt text. The AI drafts it; you read it against the final image. The alt text guide covers what belongs in it.
- The posting. You check the image at its real size on a phone, read the caption once more and press post.
A worked example
This is an editorial example with an invented business. Say you run a one-person bookkeeping practice, and you have five posts planned for the month and no photos for any of them. Your work happens at a desk, and you would rather not show client paperwork or your own face.
The brief comes back at about 90 words. It names your navy and soft yellow by hex code, a flat illustrated style, morning light, a tidy desk seen from above and a never-show list: no people, no documents with readable numbers, no money.
The five posts are tips: quarterly taxes, keeping receipts, separating business and personal accounts, a year-end checklist, and when to call a bookkeeper. The AI writes five prompts. For the receipts tip it suggests a real photo of your own receipt folder instead, which you can take in a minute without showing anything private, so that post moves to your photo list.
The other four render. One shows a calculator with a number on its screen, which breaks the no-numbers rule, so it gets drawn again. One fills the frame and leaves no room for a headline, so it gets drawn again with the desk pushed to the right. You add a short headline to each in a design tool, read each alt text line against the final picture and fix one that described a plant the image does not show. Four images and one real photo cover the month, and all five look like they came from the same desk.
Real photo, AI image or text graphic
Each post needs a real photo, an AI image or a text graphic, or software that makes them for you. The choice is about what the post has to prove.
| Choice | Who does the work | Best when |
|---|---|---|
| Real photo | You take it or pick it from your camera roll | The post shows your work, your place, your people or a result |
| AI image | The AI writes the prompt and renders it from your brief | The post explains an idea, a tip, a season or a step that a camera cannot catch |
| Text graphic | You set the words on a plain background in your colors | The post is a quote, a rule, a short list or a date |
| Software from your website | Software reads your website and makes each post with its image, and you check them | You want images that match your brand every week without writing prompts |
When a post could take either a real photo or an AI image, the real photo wins.
Boomp is software that does this job every week, without you. See ten posts made from your website, free
