Creating images with AI: how to write a good image prompt

Type "a nice coffee photo" into an AI tool and you get a nice but generic image that looks like everyone else's. The difference is in the prompt: you need to describe, piece by piece, the image you picture in your head. In this guide we will cover the building blocks of a good image prompt, 8 ready prompts for practical jobs, how to refine results step by step and how to finish the image in Canva.

What you need

• A tool that can generate images: ChatGPT, Gemini or Canva (all three have image features you can try on a free plan; limits change)

• Where the image will be used (Instagram post, YouTube thumbnail, product page…)

• Your brand colours and 2–3 example images you like, if you have them

The anatomy of a good image prompt

Keep this order in mind: what (subject), where (setting), how it looks (style, lighting, colour), how it was shot (composition, camera) and what format (aspect ratio, text in the image). You do not have to write every part, but the more parts you decide, the less is left to the AI's guesswork.

Where to generate

Burak uses it

ChatGPT

An all-round chat assistant for writing, ideas, summaries and everyday questions.

Try it ChatGPT (opens in a new tab)

ChatGPT: Just describe the image in chat; you can upload a photo to edit, or select an area of a generated image and change only that part. It is strong at following instructions and at text in images.

Gemini

A practical assistant for research and summaries that works closely with Google apps.

Try it Gemini (opens in a new tab)

Gemini: The image creation tool in the Gemini app (powered by Google's image model known as Nano Banana) can generate from scratch or edit a photo you upload, step by step through conversation. It is good at successive edits on the same image, like "now change the background, now make the light evening light".

Burak uses it

Canva

AI-assisted design tool for social graphics, presentations and more.

Try it Canva (opens in a new tab)

Canva: Generate images from text right inside your design (Magic Media / Dream Lab), brush over an area and replace it with Magic Edit, extend an image to a new aspect ratio with Magic Expand, and cut out a product with the background remover. Some features are on paid plans.

Midjourney: Especially strong for artistic, aesthetic images; it is now used through the web app at midjourney.com. You set the aspect ratio with a parameter at the end of the prompt, like --ar 16:9, and can use reference images for style or character consistency.

8 ready prompts for practical jobs

Fill in the brackets with your own details. The prompts work as they are in ChatGPT and Gemini; in Canva and Midjourney you may need to shorten them.

1. E-commerce product photo

Create a realistic product photo. Product: [name, colour, material, size]. The product sits in the exact centre on a plain [light grey / cream] surface. Lighting: soft studio light with a subtle, natural shadow. Camera: close to eye level, slightly above, the whole product in frame. No other objects in the background. Square (1:1) format. Do not change the product's shape, colour or logo.

If you have your own product photo, uploading it and saying "keep this product exactly as it is, only change the background and lighting" usually gives a more reliable result.

2. Social media post background

Create a background image for an Instagram post. Theme: [e.g. autumn sale / new course announcement]. Style: [minimal / warm / energetic]. Colour palette: [your brand colours, e.g. navy and mustard]. Keep the top half clean and empty; I will add text there later. No text in the image. Vertical 4:5 format.

3. YouTube thumbnail concept

Create a 16:9 thumbnail concept for a YouTube video. Topic: [topic]. On the right: [what is in the scene, e.g. a laptop on a desk and a surprised cartoon character]; leave the left side empty for text. High contrast, vivid colours, a simple composition that still reads at small sizes. No text in the image.

If you want your own face on the thumbnail, placing a real photo of yourself over this background in Canva looks more natural.

4. Logo and icon moodboard

I want to explore logo directions for [business name and what it does]. Create a single moodboard image with 9 different logo mark (icon) concepts in a 3x3 grid. Each should try a different approach: letter-based, abstract shape, based on [an object related to the industry]. Flat vector style, at most 2 colours, white background. No text.

5. Blog or carousel illustration

Create an illustration for a blog post about [topic, e.g. staying focused while working from home]. Scene: [what should be shown]. Style: flat, modern editorial illustration, soft textures, limited palette ([2–3 colours]). Keep characters simple, without facial detail. Horizontal 16:9. I will create more images for this series, so keep the style consistent.

For a carousel, once the first image works, continue in the same chat with "same style and colours, this time [new scene]".

6. Portrait style from your own photo

The person in the photo I uploaded is me. Keeping my facial features, hair and expression, turn this photo into a [professional LinkedIn headshot / warm-toned magazine portrait]. Background: [plain dark grey / blurred office]. Lighting: soft, from the front. Clothing: [e.g. a navy blazer]. Do not slim my face or add beautifying effects; keep it natural.

Only do this with your own photo or a photo of someone who has given clear permission.

7. Mockup

Place the [logo / design] I uploaded onto a realistic [white T-shirt / coffee mug / book cover / phone screen] mockup. Do not change the design; apply it cleanly with natural perspective. Setting: [e.g. a wooden table in natural light]. Square 1:1.

8. Infographic-style image

Create a simple infographic about [topic]. Title: "[short title]". Include 4 steps, each with an icon and a label of at most 3 words: 1) [step 1] 2) [step 2] 3) [step 3] 4) [step 4]. Flat vector style, white background, [your brand colour] as the accent colour. Vertical 4:5. Use the text exactly as written.

Always check the text on infographics. If there is even one wrong letter, generating the image without text and adding it in Canva is faster.

Refining the result: steer, do not start over

The first image is rarely the final one. I usually work like this: get the composition right first, then fix details. Change one thing at a time; ask for five changes at once and you will not know what worked.

1. Say what you like: "The composition and lighting are great, keep them."

2. Ask for one change: "Only make the cup dark green."

3. Name the problem: "The background is too busy, remove the objects on the shelves."

4. Rewrite if you are stuck: If something is still wrong after three tries, rewrite the prompt from scratch with the elements you liked, more clearly, in a new chat.

When you get an image you like, turn it into a reusable recipe with this prompt:

Write a detailed image prompt that would let me recreate this image, using these headings: subject, setting, style, composition and camera, lighting, colour palette and aspect ratio. Then turn it into a template with brackets that I can reuse for different subjects in the same style.

Finish in Canva

An AI image is usually raw material. To make it ready to post:

• Upload the image to Canva and place it in a design with the right dimensions (post, story, thumbnail).

• Add text in Canva, using your brand fonts and colours.

• If the image is too tight, extend it with Magic Expand; remove unwanted objects with Magic Edit or the eraser.

• For product images, remove the background and use your own.

• Check how it looks at small size (on a phone), then export.

Common mistakes

1. One-word prompts: Use the building blocks instead of "coffee photo".

2. Not stating the aspect ratio: Cropping a square image into a vertical one ruins the composition.

3. Putting long text in the image: More text means more errors; add it in Canva.

4. Contradictory instructions: Requests like "minimal but very detailed" give indecisive results.

5. Starting over every time: Refining an image you like in the same chat is much faster.

6. Not checking details: Hands, text, logos and product details break most often; zoom in before you post.

Next step

Today, pick the prompt closest to your needs, fill it in and do two rounds of refinement. Then use the BONUS prompt to create your own template and save it somewhere. For more ready prompts and guides, browse the resources; to hear when new ones come out, join the newsletter.