You describe what you need in words and it creates it for you: an image for a post, the header of an article, a product mockup, an illustration for a presentation. Just like with text, the magic is in the detail of the description — and in the fact that your Agent knows your brand and what you want it for.
This guide is about going from "it made me something generic and stock-photo-ish" to "this fits perfectly with what I needed."
1. The anatomy of a good image brief
Think like a director of photography giving instructions to someone. These are the levers:
| Lever | The question it answers |
|---|---|
| Subject | What appears in the image? |
| Composition / framing | Top-down view? Close-up? Where do I leave room for text? |
| Style | Realistic photo, illustration, 3D, watercolor…? |
| Light and mood | Warm natural light, dramatic, neon, soft? |
| Color | Which palette? Your brand colors? |
| Format | Square, horizontal, vertical? For which platform? |
| What it's for | Post, header, ad, slide background? |
| What to avoid | No text, no people, no visible logos? |
The leap in quality, in one example:
❌ Weak: "Make me an image for a post."
✅ Detailed: "Create an image for an Instagram post (square, 1:1) about productivity. I want a minimalist desk seen from above (top-down): a laptop, a notebook, a coffee and a little plant, on light wood. Realistic photography style, soft natural morning light coming through a window, palette of warm, neutral tones. Leave the top half clear so I can put text on top. No people or visible logos."
Every sentence is a decision that, if you don't make it, it makes for you at random.
2. What you're actually going to use it for
Forget "draw me a cat." These are the uses that genuinely save you time and money on stock photos:
Blog header (16:9): "A horizontal 16:9 image to head up an article titled 'How to automate your invoicing.' Modern flat illustration style, conceptual: documents and invoices flowing toward a gear, a sense of order and automation. Corporate palette in blues with an orange accent. Clean space on the left to overlay the title."
Product mockup: "A clean 3D render of a natural cosmetics box on a neutral background, soft studio light, subtle shadow, three-quarter view. Cream background, lots of air around it. Premium look, like a catalog photo."
Illustration for a presentation: "An isometric illustration of a small team working with AI assistants, modern flat style like a corporate slide, flat colors, no flashy gradients. White background so I can drop it into a slide."
Visual for an ad (9:16 story): "A vertical 9:16 image for a story. A steaming cup of coffee on a desk at dawn, warm tones, soft focus (blurred background). It should convey 'start your day with an edge.' Leave the bottom third free for the call-to-action button."
3. The vocabulary of styles (your palette for asking)
Naming the style changes everything. Keep these handy:
- Realistic photography — to make it look like a real photo.
- Flat illustration — modern, clean, web/app style.
- 3D render — products, objects, a premium look.
- Isometric — technical/corporate scenes with perspective.
- Watercolor / hand-painted — warm, human, editorial.
- Line art / minimalist — icons, simple strokes, lots of air.
- Retro / vintage — nostalgic, '70s-'80s.
- Abstract gradients — backgrounds for slides or banners with no subject.
Trick: if you don't know which style you want, ask it for options: "Make me the same scene in three styles —realistic photo, flat illustration and watercolor— and I'll pick."
4. Format and aspect ratio (nailing the size)
Asking for the right aspect ratio saves you from ugly crops. Always say it:
| For… | Aspect ratio | How to ask for it |
|---|---|---|
| Instagram post / feed | 1:1 | "square" |
| Story / Reel / TikTok | 9:16 | "vertical, story format" |
| Blog header / YouTube / X | 16:9 | "horizontal, widescreen" |
| Banner / email header | ~3:1 | "wide and narrow, banner-style" |
| Portrait / vertical product | 4:5 | "slightly vertical" |
5. The nuance: the same image, several registers
As with tone in text, a change of style or light transforms the result. Ask for it and compare:
"Take the desk scene from before and make me four versions: (1) realistic photography with morning light; (2) minimalist flat illustration; (3) clean 3D render; (4) warm watercolor. Same framing in all of them. I want to see which one fits my brand best."
Mood variant: "…now the same realistic photo but changing the light: one version with warm dawn light and another with cool, modern office light. Let's see which one better conveys 'focus and calm.'"
6. Brand and consistency
Don't let it look like it came from just anywhere: make it look like yours.
"Use my brand palette: orange #FF5733 as the accent and a cream background. Keep the same flat illustration style as the last image you made me, so the series stays consistent."
"Make me a set of 3 matching icons (same line weight, same colors) for the three services on my website: consulting, training and support."
7. Text inside images (read this)
Image generation chokes on text: it usually comes out warped or with typos. So:
- If you need an image with text, ask for it without text and leave the gap ("leave the top area clean so I can add the title"). You add the text afterward in Canva/PowerPoint.
- If you still want it to try, make it a single short word and tell it clearly: "one big word that says SALE, no other text at all."
8. What changes everything: it knows your context
Your Agent doesn't generate in a vacuum — it knows what you're working on:
"Make me an image to go with the article you just wrote about bank reconciliation, one that matches its tone."
"A header for this week's newsletter, in the same visual style we've been using."
"Illustrate the Clinica Jane success story with something understated and clinical, without falling into the typical stock photo of smiling doctors."
9. Iterating: the first one isn't the final one
Sculpt the result through conversation:
- "More minimalist, there's too much background noise."
- "Swap the plant for a cup of coffee."
- "Same framing but horizontal 16:9 for the blog header."
- "Remove the people in the background."
- "Make me 4 variations of this and I'll compare them."
- "I like this one: now give it to me in a vertical story version too."
10. Pro-level tricks
- Describe it like a photographer: framing, type of light, even the "lens" ("like with a 50mm, blurred background").
- Ask for options, not just one: "give me 4 and I'll pick" almost always beats nailing it on the first try.
- Chain it with the text lesson: "Write me the LinkedIn post about this and make me the image to go with it, matching."
- Give it a reference: "something with the feel of this image I'm sending you" (send it an example).
- Always think about the gap for text from the very first prompt if the image is going to have a title on top.
11. Common mistakes (and how to avoid them)
| Mistake | What happens | Fix it like this |
|---|---|---|
| Vague prompt | You get something generic and stocky | Subject + style + light + what it's for |
| Not saying the format | You get a square one when you wanted horizontal | Always state the aspect ratio |
| Cramming in lots of text | It comes out warped | Leave the gap and add the text yourself |
| Asking for logos/brands/celebrities | Rights problems | Describe concepts, not registered trademarks |
| Settling for the first one | You make do | Ask for variations and iterate |
12. A practical note
The images you generate are for your own use, but use your head: don't ask for registered brand logos, copyrighted characters, or the faces of real, well-known people. For important uses (an ad you're going to put budget behind), review it carefully before publishing.
Try it now
Copy it, adapt it and paste it to your Agent:
"Make me an image [format: square/horizontal/vertical] for [use] about [topic]. I want [subject and scene], [realistic photo/flat illustration/3D] style, [warm/soft/dramatic] light, [colors] palette. Leave [a gap for text / no text]. No logos or people. And if you like, give me 3 variations."