Issue 1 · 19 November 2026 · 3 min read
The 10-step order I use for every AI image prompt
Most AI images look generic because the prompt is written in the wrong order. Here is the exact build order I use for every image, in any tool, with prompts you can copy.
By Lynn Noppen
If your AI images look a little flat, a little stock, a little "AI", the problem usually isn't the tool. It's the order you write the prompt in.
Image models read a prompt from the start. What comes first gets the most weight. So when a prompt opens with "beautiful, cinematic, moody, aesthetic", the model spends its attention on vibes before it even knows what it's drawing.
I write every image prompt in the same order, whether I'm in ChatGPT, Midjourney, Nano Banana or Canva's AI tools. I call it the Sought After Studio Prompt Method. Here it is, step by step.
The build order
Write the blocks in this order. Skip one if it doesn't apply, but never shuffle them.
- Subject. One plain clause. No adjectives yet. A ceramic mug on a wooden table.
- Spatial relationships. If there's more than one thing in the frame, say exactly where each one sits. The mug in the foreground right, a linen napkin folded beside it, a small vase behind them on the left.
- Reference role. If you upload a reference image, say what it's for: the shape, the colour, the material or the mood. Not all four.
- Material and texture. What things are made of and how the surface reacts to light. Matte stoneware with a speckled glaze.
- Colour. Name the tone, don't describe a feeling. Warm oat with a dusty terracotta rim, not "neutral".
- Setting. One short phrase. A quiet kitchen counter.
- Lighting. Direction, quality and source in one line. Soft window light from camera left, gentle falloff, no fill.
- Camera and lens. Angle, lens and depth of field. Shot at eye level on a 50mm lens, shallow depth of field, natural film grain.
- Style and mood. Three to five words, at the very end. Calm, editorial, Scandinavian.
- Negative control. Only for the thing the tool keeps getting wrong. No text on the mug.
Notice where the mood words went: last. By the time the model reads them, everything that matters is already decided.
The full prompt
Put together, the mug example reads like this:
A ceramic mug on a wooden table. The mug in the foreground right, a linen napkin folded beside it, a small vase with one dried stem behind them on the left. Matte stoneware with a speckled glaze, the napkin in soft washed linen. Warm oat mug with a dusty terracotta rim, natural oak table. A quiet kitchen counter. Soft window light from camera left, gentle falloff, no fill. Shot at eye level on a 50mm lens, shallow depth of field, natural film grain. Calm, editorial, Scandinavian. No text or logos on the mug.
Swap the subject for your own product and keep the rest. You'll see the difference straight away.
Three rules that make it work
Name exact values, not feelings
"Warmer", "better", "more natural": a model can't act on a feeling. Name the tone, the texture or the angle instead.
- Instead of good lighting, write soft light from camera left, gentle shadow underneath.
- Instead of nice colour, write warm terracotta, slightly dusty, not saturated.
- Instead of make it pop, write higher contrast, one saturated accent colour only.
- Instead of a few things on the table, write a mug left of centre, a notebook angled beside it, a pen resting on top.
Lock only what matters
Pick the two or three things that really can't change: maybe the product shape and the brand colour. Leave everything else open. When you lock everything, the result gets stiff and generic.
Fix one thing at a time
Got an image that's almost right? Don't rewrite the whole prompt. Say the one change, then lock the rest:
Change only the mug colour to deep forest green with a matte finish. Keep the camera angle, framing, composition, lighting direction and every other object exactly the same.
Your turn
Take one product photo you need this week. Write the prompt in this order, run it once, then use the fix prompt to change a single detail. Hit reply and tell me which step made the biggest difference. I read every email.
Next week: how to make AI write captions that actually sound like you.
