What image generation is genuinely good for at work, how to ask for a picture the way you'd brief a designer, where video stands today, and the two questions to ask before you use any of it in public.
Image generation is the part of AI that looks like magic in a demo and like a hobby in the office. That's because most people start by making something fun and then can't think what to use it for. Start from the other end: the pictures you already need and currently don't have.
What it's actually for at work
The honest list, in order of how often it pays off:
Slides and documents. A cover image, a visual metaphor for a section, an illustration where you'd otherwise use a stock photo of people shaking hands.
Mock-ups and placeholders. "Show me roughly what a poster for this event could look like" before you brief a real designer. Options to react to, not finals.
Simple diagrams and icons. Good for consistent, plain icons. Weaker for precise diagrams; for those, describe the diagram to a text assistant and have it draw it in code, which is how the pictures in this course were made.
Editing photos you already have. Remove a background, extend an image to a new aspect ratio, clean up a product shot. Often more useful than generating from scratch.
Social posts and internal comms. Where a decent picture beats no picture and nobody expects a photographer.
Where it's weak: text inside images (better than it was, still check every letter), exact brand consistency across many images, anything that must be factually accurate (a map, a chart, a real product), and real people's faces, which you shouldn't be generating anyway.
Where to make them
You probably already have access. ChatGPT, Gemini and Copilot generate images inside the chat. Canva has it built in alongside its templates. Adobe's tools have it inside Photoshop and Express. Dedicated tools like Midjourney produce more striking results and cost a subscription. For this course, the one inside your assistant is enough. Use the same rule as lesson 2: one tool, a month, then decide.
Brief it like a designer
The mistake is writing a prompt. The fix is writing a brief, the way you'd brief a person who can't ask questions.
Four parts. The one people forget is "use": where the picture is going decides the format and what to leave empty.
Some habits that save round trips:
One subject. "A cup on a desk" works. "A busy office with people collaborating around a screen showing a dashboard" produces six wrong things at once.
Name the style. Flat illustration, watercolour, photograph, line drawing, 3D render. Without it you get a default that looks like everyone else's.
Say what to leave out. "No text", "no people", "no logos". Models add things to fill space.
Leave room for your words. "Empty space on the right" or "subject in the lower third" means you can put a title on it.
Give the format. 16:9 for slides, square for most social, 9:16 for stories. Changing aspect ratio after the fact usually means regenerating.
Then iterate the way you did with text: "same, but warmer light", "keep the composition, make it a photograph". Three rounds is normal. Ten rounds means the brief was wrong; rewrite it.
Video, as of now
Video generation exists and improves every few months. Today, for someone whose job isn't video: short clips of a few seconds look impressive; anything longer, with a consistent character, specific actions or readable text, takes many attempts and real money. The practical uses right now are short background loops, product-style shots of simple objects, and animating a still image slightly for a post. If you need a thirty-second explainer with your actual product in it, that's still a job for a person, possibly a person who uses these tools.
A more useful "video" feature is the one that doesn't generate video: turning a meeting recording into notes, or a long video into a summary. That's text work, and your assistant does it well if you give it the transcript.
Two questions before you use any of it
Who owns it, and may I use it here? Terms differ by tool and change. Most let you use what you generate commercially; some restrict it on free tiers. If it's going on a public page or in a client deliverable, spend two minutes on the tool's terms. For internal slides, nobody will ask.
Would anyone be misled? A generated "photo" of your office, your team, or a real event is a problem even if it looks nice. A generated illustration is not. If it could be mistaken for a record of something real, label it or don't use it. Never generate a real, identifiable person.
The habit
Next time you reach for a stock photo, write a four-part brief instead. Subject, setting, style with exclusions, use with format. Three iterations maximum. If it's not there by then, the stock photo was fine.
Try this today
Find a slide or document you own that has no picture or a bad one. Write the four-part brief for a replacement. Generate it, iterate twice, drop it in. Note what the first version got wrong; that's the part of the brief you'll write better next time.
Check yourself
5 questions. Everything is in the lesson above. Get them all right and the lesson counts as done.
1. Which of these is the strongest everyday use of image generation at work?
Illustrations for slides and documents are the reliable win. Real people and factual images are exactly where it fails or shouldn't be used.
2. Your generated image has a clumsy caption baked into it. Which brief line would have prevented that?
Models fill space with things you didn't ask for. Saying what to leave out is part of the brief.
3. You've done ten rounds of "a bit different" and it's still wrong. The most likely problem is:
Three rounds is normal. Ten means the brief doesn't pin down the subject, style or use. Rewrite it instead of iterating.
4. What's a realistic use of video generation today for a non-specialist?
Short, simple clips work. Long, consistent, specific video still takes many attempts and real money.
5. Which of these should you not generate, regardless of tool terms?
Real identifiable people and images that could be mistaken for records of real events are off the table. Illustrations are fine.