The anatomy of a Midjourney prompt
Midjourney parses words, not intentions. A prompt that works always follows the same order, from the load-bearing parts to the decorative ones:
- Subject and action. What is in frame and what it is doing. "A keeper climbing a spiral staircase with a lantern" beats "a lighthouse".
- Concrete details. Materials, colours, textures, era. Vague adjectives ("beautiful", "stunning") carry no information.
- Style. Visual reference, medium (photography, watercolour, 3D illustration) and period.
- Lighting. The single biggest lever: golden hour, studio light, neon, backlight, fog.
- Camera. Shot type and lens: 35 mm, portrait, wide angle, macro, top-down.
- Parameters. Aspect ratio, stylize level, model version and negatives.
Parameters worth knowing
| Parameter | What it does |
|---|---|
--ar 16:9 | Aspect ratio. Landscape for scenes, 9:16 for mobile, 1:1 for product. |
--stylize 0-1000 | How much artistic freedom you grant. Low sticks to your text, high beautifies. |
--chaos | Variety across the four proposals. High explores, low is predictable. |
--no | Negatives: what you do not want (watermark, text, mangled hands). |
--version | Model version: each one shifts look and fidelity. |
Repeated mistakes
- Stacking fifteen adjectives. Attention gets spread and you get an undefined image.
- Asking for readable text. Image models are bad at typography: generate without text and add it later.
- Forgetting the light. It is the variable that most separates amateur from convincing.
- Not declaring the style. Without a medium or reference, the model picks one for you and drifts between runs.
- Dumping everything in random order. Order matters: what comes first weighs more.
Example of a generated prompt
The same prompt works in DALL·E, Flux or Stable Diffusion, though each model has its quirks: the image prompt guide covers what changes between them.
Images that also sell
When the prompt is for a product shot, a cover or an ad, the first question is not style but function. A catalogue image needs a clean background and reproducible light; a cover needs negative space for the headline. Saying that in the prompt saves twenty generations and an editing session.
Keep going with the generator
Frequently asked questions
How do you write a Midjourney prompt?
Order the information: subject and action, concrete details, style, lighting, camera, then parameters at the end. The more specific the subject and the light, the closer the result gets to what you pictured.
Which Midjourney parameters matter most?
Aspect ratio (--ar), stylize level (--stylize), chaos (--chaos) to explore variants, negatives (--no) and the model version. Those five control most of the outcome.
Why do my images all come out looking the same?
Because the prompt barely changes: same light, same style and same camera give near-identical results. Change the light and the shot type between variants to get genuinely different options.
Can I reuse the same prompt in DALL·E or Stable Diffusion?
Yes, the structure holds, but parameters differ: DALL·E ignores --ar and Stable Diffusion uses per-word weights. Keep the text, adjust the parameters.
Does Midjourney render text inside images?
It tries, but it usually comes out malformed. Generate without text and add it in an editor — faster and cleaner.
Is this prompt generator free?
Yes — free, no signup, up to 8 prompts per hour per IP.