MidjourneyPrompting basicsParameters & settings
Midjourney Prompt Generator: Build Prompts That Work
The most reliable Midjourney prompt generator is a fixed structure you fill in: subject, medium, setting, light, colour, mood and composition, followed by parameters such as --ar, --raw and --s. Midjourney's own Describe tool and Conversational mode can suggest wording, but you get better, repeatable results by choosing each slot deliberately and keeping the prompts that work.
Search for a Midjourney prompt generator and you will find dozens of pages with a text box and a button. Most of them stitch your words into a long list of adjectives, sometimes with parameters Midjourney no longer accepts. This page does something different and, we think, more useful: it gives you the structure those tools are trying to imitate, the vocabulary for each part, a correct parameter cheat-sheet for the current default model, V8.2, and the built-in tools Midjourney itself offers for writing prompts. There is no button here on purpose. A generator you understand beats one you have to trust.
Why most generators disappoint
Midjourney's own advice is that short, clear prompts usually work best, and that long lists of instructions can confuse the model. Automatic generators tend to do the opposite: they pad a simple idea with twenty style words that compete with each other, so the result is muddy and every image looks like every other generated image. Common problems:
- Stacked quality words. "Masterpiece, ultra detailed, 8k, award-winning" adds nothing a choice of medium and light would not do better.
- Instructions instead of descriptions. "Please make the cat orange" works worse than "orange cat" in a normal prompt.
- Out-of-date parameters. Many still output
--q 2, which V8.x does not support, or--cref, which has been replaced by the Edit model. - Negatives in the text. "No people" in the prompt tends to add people. Exclusions belong in
--no.
The seven-slot prompt structure
Midjourney's Prompt Basics lists the details worth deciding: subject, medium, environment, lighting, colour, mood and composition. Use them as slots, fill only the ones that matter for this image, and keep the order. A useful template:
[medium] of [subject doing something], [setting], [light], [colour], [mood], [composition] --parameters
- Subject. Who or what, with a number rather than a plural ("three cats", "a flock of birds") and specific words ("enormous" rather than "big").
- Medium. Photograph, oil painting, risograph print, pencil sketch, 3D render. This one word changes more than any other.
- Setting. Where: a salt flat, a tiled kitchen, a cave, the deck of a ferry.
- Light. Overcast, low winter sun, neon, softbox, candlelight.
- Colour. Muted, duotone, pastel, monochrome, acid green.
- Mood. Calm, tense, playful, lonely.
- Composition. Close-up, wide shot, bird's-eye view, centred, rule of thirds.
Our prompt structure guide goes deeper on order and weight.
Word banks for each slot
Most weak prompts are weak because the writer did not have the word. Pick one or two from a row, not five.
| Slot | Words that change the image |
|---|---|
| Medium | 35mm photograph, medium-format photograph, gouache, ink wash, linocut, cyanotype, ukiyo-e, risograph, cut paper, claymation, isometric 3D render |
| Light | overcast diffused daylight, golden hour backlight, rim light, hard noon sun, window light from the left, softbox, sodium street light, candlelight |
| Camera and lens | 24mm wide angle, 85mm portrait lens, macro, tilt-shift, shallow depth of field, long exposure, motion blur |
| Composition | extreme close-up, three-quarter view, low angle, top-down flat lay, negative space, symmetrical, off-centre subject |
| Colour | muted earth tones, duotone, pastel, monochrome, complementary colours, saturated primaries, sepia |
| Era and place | 1970s, 1920s poster, Scandinavian interior, Brutalist, Mediterranean, mid-century |
| Mood | serene, eerie, nostalgic, triumphant, intimate, melancholic |
Midjourney's Art of Prompting page shows the same idea with one subject: change only the medium, the decade or the colour and the whole image changes. For the camera row, our guides to lighting and camera angles have fuller lists.
Parameter cheat-sheet (V8.2)
Parameters go at the very end, each with two hyphens and a space before them, no punctuation.
| Setting | What it does | Start with |
|---|---|---|
--ar | Image shape | 4:5 portrait, 16:9 wide, 1:1 square |
--raw | Removes default styling | On for photography and products |
--s | Midjourney's own styling, 0 to 1000 | 50 literal, 100 default, 300+ expressive |
--c | Variety within the grid, 0 to 100 | 20 to 40 when exploring |
--no | Things to leave out | --no text, watermark |
--sref / --sw | Style reference and its strength | One code, --sw 100 |
--iw | Weight of an image prompt, 0 to 3 | 1, then 0.5 or 2 |
--seed | Repeatable starting noise | Only while testing one change |
--hd | 2048-pixel images | For finals, not exploration |
--draft | Cheap, fast exploration | First runs of a new idea |
Every parameter has its own section in Midjourney parameters explained, and the settings panel lets you make any of them a default.
Built-in tools that suggest prompts
Midjourney ships three honest helpers. None of them reproduces an existing image exactly, and Midjourney says so.
- Describe. Drag an image onto the Imagine bar, or right-click one of yours and choose Describe Image. Midjourney returns four suggested prompts, longer and more detailed for V8.1 and V8.2. Each run gives different suggestions. Use it to discover style words, not to copy a picture.
- Conversational mode. Describe your idea in ordinary language, by text or voice, and an AI writes the prompts for you. It works with V7 and V8.1. You give up direct control of the wording, so copy the prompt it wrote and edit it once you like a result.
- Suggest Prompt in the Editor. In the Edit tab, Suggest Prompt uses Describe to write a prompt from the image you are editing.
Recovering the prompt of an image
A different question is "what prompt made this picture?" Describing an image with AI only invents a new prompt. The real one survives only in the file's metadata, if at all. PNGs downloaded from midjourney.com often still carry their prompt, parameters and Job ID; images saved from Discord, social media or screenshots usually have lost it. Our free Image to Prompt tool reads what is really stored in a file, in your browser, and says plainly when nothing is there.
How to iterate like a generator
- Write the shortest prompt that names subject, medium and light.
- Run it in Draft mode or at a low cost to see the direction.
- Change one slot per run, with a fixed
--seedif you want a clean comparison. - Add parameters last: ratio, Raw, stylize, then a style reference.
- Save the exact prompt and parameters next to every image you keep. A prompt is only reusable if you can find it again.
For text in images, put the words in double quotation marks ("with the words "open late" on the sign"); short Latin-alphabet phrases work best, and Raw or a lower stylize helps.
Try it
risograph print of a cyclist crossing a bridge at dawn, river mist, low winter sun behind, two-colour palette of teal and coral, calm, wide shot with lots of negative space --ar 3:2 --s 150
Now generate your own version by swapping one slot at a time: change "risograph print" to "35mm photograph" and add --raw, or replace the light with "sodium street light at night". Each swap shows what that slot is worth.