The prompt formula
Lesson 3 of 13 in the course “AI Images and Slides”. Free, with animations and a quiz.
Six parts of a prompt
A good image prompt answers six questions. What’s in the frame and what it looks like. What’s happening. Where. What the light is. How it’s shot: close or far, from which angle. And what style and format. Liya writes: a candle in a tin, cream wax, kraft label, burning on a wooden windowsill in the morning, soft light from a window on the left, close-up, blurred background, realistic photo, 4:5. Copy the template below and fill in yours.
Vague vs precise
Compare two prompts. “A beautiful candle for a post” leaves every choice to the model: which candle, where, what light. So each time you get a new candle that isn’t yours. A prompt built on the formula narrows the choice, and all four options look like Liya’s candle. Picking among similar ones is easy, and they need only small edits. Length alone doesn’t matter: concrete details do, not words like “masterpiece” and “8K”.
Talk like a photographer
Models learned from captioned photos, so they understand photographers’ language. Close-up: the object fills almost the whole frame. Top-down, or flat lay: the camera looks down at a table, the classic way to shoot arrangements. Soft window light gives gentle shadows without glare. Golden hour: warm low light before sunset. A blurred background removes clutter and leads the eye straight to the product. Two or three such words do more than ten adjectives.
Set the format up front
If you don’t set a format, many services draw a square, and cropping later cuts off the top or bottom. So write the format in the prompt or pick it with a button. 1:1 is a square. For a feed post use 3:4 or 4:5: since 2025 the Instagram profile grid shows posts in 3:4. 9:16 is for full-screen stories and Reels. 16:9 is for presentation slides and YouTube thumbnails. Liya makes her poster in 3:4 and her product carousel in 4:5.
Edit, don’t restart
ChatGPT, Gemini and Alice AI follow the conversation. You don’t need to rewrite a long prompt each time. Pick the best option and say what to change: “same, but warmer light”, then “add a eucalyptus sprig on the right”. One edit at a time works better than five at once: the model gets less confused and doesn’t touch what already works. Three steps later Liya has her product card.
Say what you want, not “no”
Liya needs an empty street as a backdrop, but the model keeps drawing passers-by. Many models handle negation poorly: the word “people” in a prompt can actually nudge them to draw people. It’s safer to describe what should be there: “an empty street at dawn, closed cafés”. If a service has a separate field for exclusions, put the unwanted things there: Midjourney has the --no parameter for this. In chats a follow-up helps: “remove the people from this picture”.
Check yourself
- Liya wrote: “a candle in a jar on a table, realistic photo, 4:5”. What’s missing from the formula?
Correct answer: Light and shot: where the light comes from and how it’s framed. What, where, style and format are there. The model will choose the light and framing itself, at random. - Liya is making a picture for stories. Which format should she use?
Correct answer: 9:16. Stories open full-screen on a phone, which is a 9:16 vertical. - Liya needs a street with no one on it, but the model draws passers-by. What should she try first?
Correct answer: Describe the scene: “an empty street at dawn, closed cafés”. Many models struggle with “no”. An image of an empty street shows what to draw without suggesting extras.