How to Create Images With AI for Slides, Docs, and Posts

Every slide deck has that moment: you need a picture, the stock photo site offers the same six smiling handshake people it offered in 2015, and the clipart makes everything look like a school project. AI image generation ended that era. The image tools inside ChatGPT, Gemini, and Canva now make custom illustrations good enough for real work, free, in under a minute.

What most office workers are missing is not access; it is the three habits that separate a usable work image from an AI-looking one: describe like a brief, keep a style recipe, and inspect before you ship.

Where AI images belong at work (and where not)

Great fits: concept illustrations for slides, blog and intranet headers, icons and simple diagrams, mockup imagery for proposals, and internal posters. Poor fits: photos of your actual product (generate fiction and customers will notice), real people, charts with data (make those from the data itself), and anything a regulator might read. The dividing line is honesty: illustration decorates an idea, but a generated image posing as a real photo of a real thing is a lie with better lighting.

Describe it like a design brief

The prompt formula that works for images mirrors the one for text, with craft words added:

Create an illustration for [where it will be used]. Subject: [what is happening, concretely: a person sorting envelopes into trays, not "productivity"]. Style: [flat vector illustration / watercolor / 3D render / line art], [minimal / detailed]. Colors: mostly [your brand colors or a mood], accents in [color]. Mood: [friendly / serious / energetic]. Composition: [lots of empty space, subject on the left]. No text or letters anywhere in the image.

Two details do heavy lifting. Concrete subjects beat abstractions, because "teamwork" produces handshake soup while "three people assembling one oversized puzzle" produces a picture. And the no-text rule exists because AI still misspells any words it draws, which is the fastest way to make a slide look broken.

The style recipe: consistency is what looks professional

One good image is luck; five matching images is a brand. The trick is a reusable style block:

Help me write a reusable style recipe for my work images. Ask me about my brand colors, the mood we want, and where images will appear. Then produce one paragraph I can paste into every image prompt, covering style, palette, background, level of detail, and what to always avoid. It should be specific enough that ten different subjects come out looking like one family.

Save the recipe in your AI project next to your writing style guide, and every deck you make from now on looks art-directed. This blog runs on exactly that trick: one style paragraph, every image, same family.

Inspect before you ship

AI images fail in details: seven fingers, garbled letters on a sign in the background, objects merging where they should not. Zoom to full size and check hands, edges, text-like shapes, and physics. Then ask for targeted fixes rather than rerolling everything: "same image, but fix the hand on the left and remove the floating object near the window". Modern editors apply the change without redrawing the parts that already work.

Rights, disclosure, and common sense

Check your company's policy before using generated images in public materials, since some industries restrict them, and commercial terms vary by tool and plan. Never generate real colleagues or public figures, which lands you in deepfake territory with your own hands. For internal slides and drafts, disclosure is overkill; for published marketing, follow the same judgment you would apply to AI text.

A picture per idea, in your palette, in a minute: it is one of the most immediately satisfying AI skills to learn, and it makes your presentations better on the same afternoon. Free courses for the rest of the toolkit live here.