Prompts for blog post illustrations in Midjourney and DALL-E
A post illustration does two jobs: hold attention in a feed and avoid looking like stock. Here is the prompt formula and templates for common post types.
A five-part prompt formula
The structure that works: subject, action or state, environment, style and technique, frame parameters. Write it in English, since both Midjourney and DALL-E parse English descriptions more precisely: 'flat vector illustration of a small business owner reviewing website analytics on a laptop, warm office interior, muted teal and sand palette, clean geometric shapes, no text, wide horizontal composition'.
Part by part. The subject, 'small business owner', is a specific person instead of an abstraction. The action, 'reviewing analytics', matters because without it the model draws someone staring blankly at the camera.
The environment sets mood and removes the default white background. Style is the part that makes a blog recognisable: pick one and keep it across posts.
Templates by post type
Technical breakdowns: 'isometric illustration of connected server blocks and data flow arrows, minimal line work, two-colour palette, dark background, no text, 16:9'. Isometric art reads well at thumbnail size and needs no faces.
Guides and how-tos: 'flat illustration of a checklist and a hand marking items, soft shadows, limited palette, centred composition, no text, square format'. Lists and checklists are instantly legible visually.
Market or region posts: 'abstract illustration of a modern city skyline with geometric ornament motifs, warm sunset palette, flat vector style, no text'. Ornamental motifs give local grounding without the usual clichés.
Why every prompt says 'no text'
Both models try to add lettering and almost always mangle it: letters warp, and in Russian or Uzbek you get unreadable character soup. It is simpler to ban text in the prompt and overlay the headline in an editor.
If you need a headline inside the image, reserve room for it: add 'large empty space on the left third of the frame'. That is more reliable than hoping for a usable default composition.
One consistent style for the whole blog
Fix three things and repeat them in every prompt: a palette of three or four colours, a technique (flat vector, isometric, 3D clay, line art), and a level of detail. Without that the blog looks like a pile of unrelated images.
In Midjourney it is convenient to hold this in a style parameter and a reference image: generate one you like and use it as the visual anchor for the rest. In DALL-E, repeat the style description verbatim.
Keep prompts in a single document organised by post type. After twenty articles you will assemble an illustration in two minutes rather than half an hour.
Local context without clichés
Asking for 'Uzbekistan' almost always produces a postcard: ornaments, domes, carpets. That is off-tone for a business blog. Local flavour comes across better through palette and a geometric ornament used as a background pattern than through architecture.
If you need people in frame, describe clothing and setting in neutral business terms: 'modern office', 'business casual'. Ethnic qualifiers are handled crudely and produce caricature.
Inspect every image for details: extra fingers, warped laptop screens, unreadable charts on monitors. At feed thumbnail size these still show.
Preparing the file for publication
Generated images are usually heavy and square. Crop to the format you need, compress to WebP and keep the file under 150–200 kilobytes: page speed beats extra detail.
Always write a meaningful alt attribute in the page language: it serves accessibility and gives search engines another signal. Rename the file to something meaningful in Latin characters instead of the generator's random string.