AI Central

Better AI Images in 7 Steps

Download
AI Central
Review AI summary

A short image prompt fails for the same reason a one-line design brief fails: every gap gets filled with a cliché. AI Central's Better AI Images in 7 Steps replaces the one-liner with a brief written in fields, scene, style and format first, then the edits, characters, brand style and text that make images reusable.

What a short prompt gets you

The guide's opening example asks for an image for a post about a new invoicing app. Back comes a smiling woman at a laptop, gold coins floating around her, a blue gradient, and the headline spelled INVOCING MADE EASY. That is both failures in one image: stock-photo clichés and misspelled words.

Generic and misspelled

Steps 1 to 3: a new image from scratch

Step one describes the scene the way a camera would see it, in four fields: subject with details, action, setting and purpose. The example is a finance manager in her 40s, reading glasses pushed up on her head, ticking the last invoice on a printed list in a small office in the late afternoon. The prompt also names the post it is for and rules out screens with fake text.

Step two sets the look before the AI picks one for you: medium, light, camera and mood. The example asks for a documentary photo with natural colours, not glossy stock, soft window light from the left, eye level with shallow depth of field, and a mood of relieved and quietly proud.

Name the look, or the AI picks a cliché

Step three frames the image for the channel: 1:1 or 4:5 for feeds, 16:9 for slides, one subject placed off-centre and an empty area for your headline. The example keeps the left 40% of a slide image as a plain wall, then asks for a 4:5 version of the same scene for the feed. To change the shape, extend the canvas instead of cropping.

Step 4: edit a photo you already have

Editing follows one rule: change one thing per message and protect what must not move. Upload a sharp photo you have the rights to, select the area if your tool allows it, and name what stays fixed. The example moves a ceramic mug onto a light oak café counter in morning sun, and keeps the mug's shape, colour, glaze and logo exactly as in the photo.

Steps 5 and 6: the same look in every post

A recurring character needs one detailed description and one good reference image. You save both, attach the image for every new scene, and change only the scene, never the description. The guide's mascot, Pip, a round blue robot with one antenna and a small orange scarf, stays the same while the scene moves to a starting line for a launch-week post.

Brand style works the same way. Collect 3 to 5 images you already like, ask the AI to describe their shared style, add your colour codes and what to avoid, and paste that style block into every prompt. The example names three hex colours, allows no other colours, and lists what to avoid: gradients, 3D effects and stock-photo people.

One style block, every image on brand

Step 7: words and logos in the image

Text inside an image stays short and exact. Quote the exact words in quotation marks, keep them to a headline, and give the font style, size and position. The example prompt asks the AI to spell every word exactly as written, allows no other text, and leaves the bottom right corner empty so you can add the real logo yourself afterwards.

Choose in words before you generate

The pro tip is to let the chat write the prompt. Ask for 3 visual directions in one line each and tell it not to generate yet, pick one, then ask for the full prompt with scene, style, light and format. Comparing ideas in words is faster than generating images you throw away.

What to watch out for

  • Use only photos you have the rights to.
  • Change one thing per message when you edit, and say what must stay the same.
  • Check every letter of any text in the image before you publish.
  • Add your real logo after generation instead of asking the AI to draw it.