Midjourney prompt generator: from image to /imagine

Midjourney reads a prompt as a set of weighted ideas rather than a sentence to be parsed word by word. Words near the start carry more weight than words at the end, and long, chatty sentences tend to dilute each other. That is why our generator leads with the subject and setting, then adds medium, lighting, colour palette and mood as short comma-separated phrases.

Filler such as “a beautiful picture of” or “highly detailed 8K masterpiece” adds noise without adding information. Specific nouns and adjectives work far better: “weathered fisherman in a yellow oilskin” says more than “an old man”. When you edit the output, keep this in mind: add detail by replacing vague words, not by stacking extra adjectives at the end.

Everything after a double dash is a parameter, not description. The generator always adds --ar with the aspect ratio closest to your uploaded image, such as 3:2 or 9:16, so your result keeps the original framing. It also adds --v with the current model version, so the prompt behaves the same way whether you paste it into Discord or the web app.

For photographs we also add --style raw, which tells Midjourney to follow your words more literally and apply less of its own house style. Delete it when you want a more painterly, stylised interpretation. Other useful parameters you can add yourself include --stylize (0–1000, how strongly Midjourney applies its aesthetic), --chaos (how varied the four results are) and --no (things to leave out, such as --no text).

A text prompt describes an image; it never copies it. For a closer match, combine the generated prompt with Midjourney’s reference features: paste your image URL at the very start of the prompt as an image prompt, or add it with --sref to borrow its style. Lower --stylize values also keep results closer to the description.

If a detail keeps disappearing, move it earlier in the prompt. If an unwanted element keeps appearing, name it after --no instead of describing its absence in words. Midjourney doesn’t understand “without a hat” reliably, but --no hat works.

Generate the prompt, run it once, and compare the four results with your reference. Change one thing at a time: the subject phrase, the lighting phrase or a single parameter. Your last ten prompts are saved in this browser, so you can go back to an earlier version with one tap.