Midjourney Prompt Generator — V8 Parameters, Styles & SD/Flux Prompts
Build Midjourney prompts from style, lighting and camera chips with correct V8/V7 parameters, or switch to Stable Diffusion and Flux.
a lighthouse on a sea cliff during a storm, oil painting, rim lighting, low angle, muted earth tones, moody --ar 16:9 --s 250
Runs entirely in your browser. Prompts are built locally and nothing is sent to Midjourney or to us. Browse all free tools.
Found a problem, or something that could be better? We read every message and we fix things quickly.
This Midjourney prompt generator turns a subject and a few clicks into a prompt in current Midjourney syntax. Pick a medium, lighting, lens, composition, palette and mood from artist-neutral descriptors, set the parameters with sliders, and the prompt is written in the order Midjourney reads it: image links first, then the text, then the parameters. It knows which flags each model accepts, so you will not paste --cref into V8 or --oref into V6 and wonder why nothing changed.
The same picks can be written as a Stable Diffusion prompt with (term:1.2) emphasis and a negative prompt, or as the plain-sentence style Flux models respond to, and the size helper converts your aspect ratio to the pixel dimensions each model was trained on. It cannot judge how an image will look; Midjourney releases change model behaviour, so treat the descriptors as a starting point and iterate on the results.
How to use
- Describe the subject in plain words: what it is and what it is doing.
- Click descriptors in the categories you care about. Two or three strong choices work better than ten weak ones.
- Choose the model version and set aspect ratio, stylize and chaos. Open references to add --sref or image prompts.
- Copy the prompt and paste it after /imagine in Discord or into the box on midjourney.com.
- For Stable Diffusion or Flux, switch the mode, choose the model family and copy the positive and negative prompts, plus the pixel size from the helper.
Frequently asked questions
What order should a Midjourney prompt be in?
Image prompt URLs first, then the text description, then parameters such as --ar 16:9 --s 250 --v 7 at the very end. Within the text, put the subject first and the details after it, because Midjourney gives earlier words more weight. Parameters can come in any order among themselves, but they must all be after the text.
Which Midjourney version is the default in 2026?
V8.2 became the default model on 24 July 2026, after V8.1. You only need --v when you want a different model, for example --v 7 or --v 6.1, or --niji 7 for the anime-focused Niji model. This generator omits --v 8.2 unless you tick the box to add it.
Why does Midjourney ignore my --cref or --oref?
Character Reference (--cref) only works on V6 and Niji 6. V7 replaced it with Omni Reference (--oref, with --ow for strength), and V8 supports neither. If you switch versions, the tool removes the reference that version does not accept and shows a warning, so you can switch back or use --sref for style instead.
What do --stylize, --chaos and --weird do?
--stylize (0–1000, default 100) controls how strongly Midjourney applies its own aesthetic; low values follow your words more literally. --chaos (0–100) makes the four images in a grid more different from each other. --weird (0–3000) pushes toward unusual, offbeat results. Start with small steps such as --s 250, --c 15 or --w 250.
How do I write prompts for Stable Diffusion or Flux instead?
Stable Diffusion (SD 1.5 and SDXL) reads comma-separated tags and supports emphasis like (red scarf:1.3) plus a separate negative prompt for things to avoid. Flux models are guidance-distilled and ignore both weights and negative prompts, so describe the scene in full sentences. The mode switch in this tool writes each format from the same picks.
What image size should I use for SDXL and Flux?
SDXL works best at its training buckets of about one megapixel, such as 1024×1024, 1216×832 or 1344×768; other sizes tend to duplicate subjects. Flux accepts any size whose sides are multiples of 16, and around 1 megapixel is the usual starting point. The size helper picks the nearest bucket and computes Flux sizes from your aspect ratio.
Midjourney parameters by version (October 2026)
• V8.2 and V8.1: --ar, --s, --c, --w, --exp, --q, --seed, --tile, --raw, --no, --sref/--sw, --iw, --p. No Omni Reference, Character Reference or Niji. On V8.1, --q accepts only 1 or 4.
• V7: everything above plus --oref and --ow (Omni Reference). Raw mode can be written --raw or --style raw. --q accepts 1, 2 or 4.
• V6.1 and Niji 6: --cref and --cw (Character Reference), --style raw, --q 0.25/0.5/1/2. No --exp or --oref.
• Niji 7: --ar, --s, --c, --w, --seed, --no, --sref, --iw and raw mode; write it as --niji 7 rather than --v.
Midjourney adds and removes flags with each release. When a flag is rejected the job fails or the flag is silently ignored, so check the warning list under the prompt after switching versions.
How to write a prompt that works
Lead with a concrete subject and action: "an old fisherman mending a net on a pier" gives the model something to build. Then add one choice each for medium, lighting and framing; those three do most of the work. Mood and palette words refine the result, and rendering words such as "film grain" or "sharp focus" are the last layer.
Avoid stacking synonyms ("beautiful, stunning, gorgeous"), which dilute each other, and negatives inside the text ("no people"), which can add the very thing you name. Use --no people instead. If results drift from the description, lower --stylize or turn on raw mode; if all four images look the same, add a little --chaos.
Style descriptors here are artist-neutral on purpose. Describing technique ("loose watercolor washes, visible paper texture") is more predictable than a name and avoids imitating a living artist's work.
Stable Diffusion weighting and negative prompts
In AUTOMATIC1111, Forge and ComfyUI, (term:1.2) multiplies the attention on a term by 1.2; values between 1.1 and 1.4 are the useful range and anything above about 1.5 tends to distort the image. Plain parentheses (term) mean 1.1. Literal parentheses in a term must be escaped as \( and \), which this tool does automatically.
Negative prompts steer away from common failure modes such as blurry, watermark, extra fingers or bad anatomy. Keep them short and specific: a very long negative prompt competes with the positive one. SD 1.5 also reads prompts in 75-token chunks, so put the most important terms first.
Related free tools
- Prompt Template Builder — Write a prompt with {{variables}}, fill it from a form or a CSV, and export one prompt per row as text, CSV, or OpenAI and Anthropic batch files.
- Token Counter — Count tokens exactly for OpenAI models, estimate them for Claude, Gemini and Llama, see the cost, and watch the text split into tokens.
- ChatGPT to Word — Paste an answer from ChatGPT, Claude or Gemini and copy it into Word, Google Docs or email with real headings, tables and lists instead of ** and #.
- Placeholder Image Generator — Make a dummy image at any size with your colours and a label, and download it as PNG, JPG, WebP or SVG. Presets for social, ads and hero banners.