Day 10 · Use it on real work

Images and design, responsibly

Write an image prompt that produces something usable, and know your own line.

12 min readPlus one task, about 10 min

The one idea

Image tools are fast and genuinely useful for illustration. They can make a mood image, a simple concept scene, a poster background or a rough visual direction before a designer spends time on it. They are the wrong tool for anything that has to be true: a real event, a real person, a product condition or evidence of something that happened.

The useful mental model is not 'AI makes pictures'. It is 'a visual system makes a new image from a written description'. You control the subject, style, colour, mood and composition. You do not control every small detail, and the result is not a photograph just because it has convincing lighting.

Why it matters

A generated image can save time when the reader needs an illustration rather than proof. It can also damage trust quickly. A synthetic picture of a flood, a mayor, a customer or a product can be mistaken for a record of reality, especially when it appears without a label.

There are two honest questions before you use one. What will a reasonable reader think this picture claims? And have I described or labelled its machine-made origin clearly enough? Copyright is part of the same judgement. Commercial terms, training data, recognisable characters, living artists' styles and a client's brand can all create issues that a cheerful generation button does not solve.

The tools you will use

The tool

Image generation inside ChatGPT

A built-in image feature in the ChatGPT conversation where you describe a picture and receive a newly generated image to review and revise.

What matters about it here

Generate an image from a plain-language prompt in the existing chat
Request changes to composition, colour, mood or selected areas in follow-up messages
Use the conversation to keep the brief and the revisions together
Download the result for a draft, social post or internal visual concept

How it works: The service turns your written description into a visual representation and generates pixels that fit the requested content and style. It is not searching for one exact photograph, and it does not understand a scene as a camera operator would. Small text, hands, logos and exact counts can be unreliable. Follow-up edits produce another generated result; they do not make the image a factual record.
What it costs: ChatGPT's image access and daily limits depend on the current free or paid plan and can change. A free account may be enough for a few drafts; paid plans generally provide higher limits or priority access. Check the current terms for commercial use before publishing client work, and keep the prompt and source record.
Why this one: You already use the chat interface, so it is a low-friction way to learn the four parts of a useful image brief. It also lets you explain the intended use in the same thread, which helps you notice whether you are making illustration or pretending to make evidence.
If you cannot get it: Use the Google image tools introduced here or another service your organisation approves. A human stock image, a phone photograph or a commissioned illustration is often better when the picture must show a real product, place or person.

The tool

Google image tools

Google's set of image-generation features, available through products such as Gemini or ImageFX where the service is offered, for creating images from text prompts.

What matters about it here

Describe an image in ordinary language and generate several variations
Try different visual styles, layouts and moods for a concept or illustration
Use Google's account and product controls rather than adding another chat tool
Download or share a selected draft, subject to the product's current terms

How it works: Like other image generators, it predicts a visual arrangement that matches the prompt and samples a new image from that representation. It does not retrieve a verified scene from reality. Availability, editing options, watermarks and metadata differ by product and country, so inspect the specific interface and current terms rather than assuming every Google image tool behaves identically.
What it costs: Some Google image tools are free with an account and usage limits; others are tied to a paid Google plan. Check availability and commercial-use terms before a paid campaign. You can complete this lesson using ChatGPT or a free trial rather than buying a second service.
Why this one: It is a useful standalone alternative if ChatGPT image generation is unavailable, or if your work already sits in Google's ecosystem. Trying one alternative also shows you which parts are general to image generation: the written brief, composition and careful labelling.
If you cannot get it: Use the other image tool introduced here, an approved stock library or a designer. For a factual image, choose a real photograph with permission and a clear date or source instead of generating a plausible substitute.

From a vague idea to a usable visual brief

Choose illustration before you open the image tool. A picture of 'a person organising paperwork' is an illustration. A picture of 'the person who signed this contract at yesterday's meeting' claims a real event and should not be generated as if it were evidence.
Write the subject and the purpose. State what the viewer should notice, not only what objects should appear. 'A calm visual for a beginner lesson about document review' gives the image a job.

Add style, colour and mood, then composition. Composition is the part beginners omit: wide banner, subject on the left, empty space on the right for a headline, or square crop for a profile card.

Subject: a person organising a small stack of paperwork
Purpose: an illustration for a beginner lesson about careful document review
Style: clean editorial illustration, not a photograph
Colour and mood: warm cream, dark green and blue, calm and practical
Composition: wide banner, subject on the left, clear empty space on the right for text
Avoid: logos, readable fake text, real people and claims of a real event

Generate a first version, then improve it three times by changing one part each time. Ask for more empty space, a different crop or a simpler style. Do not change everything at once, or you will not know which instruction helped.
Inspect the result as a publisher. Check hands, text, logos, repeated objects and anything that could be mistaken for a real person or event. An attractive error is still an error.
Save the prompt, the date, the tool and the disclosure you intend to use. If the image is for a client, check their policy and the generator's current commercial terms before delivery.

A written prompt card flows through an arrow into a framed abstract picture with a deliberate composition. — the written brief supplies subject, style and layout before the picture is made.

See it

Compare two prompts for the same lesson header. 'Make an image about paperwork' leaves the system to choose the subject, style, crop, mood and space for the title. The result may be pleasant but impossible to place in the page.

A stronger brief says: 'Create a clean editorial illustration of a person sorting a small stack of papers, warm cream and dark green palette, calm practical mood, wide banner, person on the left, generous blank space on the right for a headline, no logos and no readable text.' It does not guarantee a perfect image. It removes several decisions that the generator would otherwise make for you.

Now change the subject to 'the mayor signing the new town contract yesterday' and the category changes. A polished result would still be an invented scene. Lighting and detail do not turn it into reporting. The first prompt asks for illustration; the second invites a reader to believe a false record.

Illustrate something of yours, and find your line

The generation is the easy half. Writing down where you would never use it is the part that makes this a lesson rather than a novelty, and the finished prompt and image are something you can keep.

Find an image tool. ChatGPT has image generation in some plans, and Google's image tools are an alternative where available. Read the tool briefing first and use only an account or service your work permits. Free access is fine.
Choose an illustration job you actually have: a lesson header, a calm background for a presentation, a concept for a service page or a diagram-like visual. Do not generate a real person, real event, product condition or piece of evidence.

Write the prompt with all four parts. Composition is the part beginners always omit, and it is the part that decides whether you can use the result. Save the prompt before you generate.

Subject: [what is in the picture]
Purpose: [where it will be used and what the viewer should notice]
Style: [flat illustration / photo-like concept / diagram / sketch]
Colour and mood: [e.g. warm, calm, blue and grey]
Composition: [e.g. wide banner, subject on the left, space for text on the right]
Avoid: [logos, readable fake text, real people or anything misleading]

Generate it, then improve it three times, changing one part each time — the same discipline as Day 3. Try composition first, then colour, then the level of detail. Keep the best version and note why it won.
Now write down where that image would and would not be acceptable to use, specific to your own situation. Include at least three things in the second line: real people, real events, product photos customers rely on, anything used as evidence, news or anything a reader would assume was a photograph of something that happened.
```
I would use an AI image for: ______
I would never use an AI image for: ______
I will check copyright, client policy and commercial terms before: ______
```
Add one line you will actually follow, and mean it. Label the image where a reader can see the label. Machine-made images should not quietly pass as photographs, and a label does not make a misleading image acceptable in the first place.
```
When I use a generated image, I will say it is generated in: ______
```

What you should notice

The prompt with composition details is dramatically better than the vague one. Saying 'leave space for text on the right' is the difference between an image you can use and an image you cannot. A visual brief is not a spell. It is a way to make the decisions visible before the tool makes them for you.

The second finding is about honesty. The line is not that AI images are bad. Illustration and evidence are different categories, and the same tool is excellent at one and dangerous for the other. If a reader might reasonably assume a picture records a person, place or event, do not generate a substitute and let the context do the lying for you.

Copyright is not solved by calling the result original. Check the generator's current terms, avoid copying a living artist's distinctive style as a shortcut, do not use protected characters or logos without permission, and follow a client's policy. Label machine-made images plainly. The label is a minimum act of honesty, not a licence to publish anything.

A framed illustration sits beside a small disclosure badge indicating that the image was machine-made. — the disclosure is part of publishing the picture, not an optional decoration.

If you have five more minutes

Ask for a diagram rather than a picture — a simple flowchart, process map or comparison layout. In real work these are often more useful than decoration. Check every label and arrow yourself; generated diagrams can look orderly while putting the wrong words or relationships in the boxes.

Tomorrow

What automation actually means, with the hype removed.

Get the rest by email

One lesson each morning at 07:00 IST, with the task written out in full so you can do it from your inbox.

This is the course, not the newsletter. Joining does not subscribe you to the daily Brief.