Image Prompt Builder
Describe your subject, pick a style, lighting and camera, and get a ready-to-paste prompt for Midjourney, Stable Diffusion, DALL·E or Flux. Runs entirely in your browser.
How to write a better image prompt
Image models respond to the same handful of ingredients: a clear subject, a visual style, the lighting, how the shot is framed, and the mood or palette. Most weak prompts are missing two or three of these. This builder makes you fill them in, then assembles them in the order each model prefers.
Model differences
Midjourney reads natural comma-separated phrases and takes parameters like --ar 16:9 at the end. Stable Diffusion responds well to comma-separated keyword lists and uses a separate negative prompt for things to avoid. Flux models generally work best with clear natural-language sentences and usually don’t use a negative prompt, so describe what you want instead. DALL·E and GPT image prefer a plain descriptive sentence and ignore parameter flags.
Frequently asked questions
Does this generate the image? No — it writes the prompt. Paste it into whichever image tool you use.
Is anything sent to a server? No. The prompt is assembled from your selections in your browser.
What is an image prompt builder?
An image prompt builder turns a few simple choices (subject, style, lighting, framing, mood and colour) into a text prompt that AI image generators understand. Instead of staring at an empty box and typing “a nice picture of a lighthouse”, you pick the ingredients from menus and the builder puts them in a sensible order for the model you use.
It works as a Midjourney prompt generator, a Stable Diffusion or Flux prompt writer, and a DALL·E prompt helper. The builder only writes the text; you copy it and paste it into your image tool of choice, which does the actual generating.
Anatomy of a good image prompt
- Subject first. Say what is in the picture and what it is doing: “an elderly fisherman mending a net on a wooden pier” beats “a fisherman”.
- Medium or style. Photo, oil painting, 3D render, anime, flat vector. This single choice changes the result more than anything else.
- Framing. Close-up, wide shot, aerial view or a lens such as 85mm tells the model where the “camera” is.
- Lighting. Golden hour, studio lighting or neon sets the time of day and the contrast.
- Mood and palette. Words like “serene” or “teal and orange” steer the overall feel.
- Technical extras. Aspect ratio, detail keywords and (for Stable Diffusion) a negative prompt.
Example prompts from the builder
Say you enter the subject “a lighthouse on a cliff at dusk” and choose cinematic film still, wide shot, blue hour, dramatic, teal and orange, and highly detailed. Here is what the builder produces for each target model.
Midjourney (with 16:9 selected):a lighthouse on a cliff at dusk, cinematic film still, wide shot, blue hour, dramatic atmosphere, teal and orange, highly detailed --ar 16:9 --v 7
Stable Diffusion / Flux: the same comma-separated list without the flags, followed by a separate line reading Negative prompt: blurry, low quality, deformed, watermark, text. For Flux, drop the negative prompt line and rewrite the list as one or two clear natural-language sentences.
DALL·E / GPT image:A cinematic film still of a lighthouse on a cliff at dusk, wide shot, blue hour lighting, dramatic mood, teal and orange palette, highly detailed.
Notice that the subject already says “at dusk”, so blue hour fits it. Picking “harsh midday sun” instead would give the model conflicting instructions, and the result usually looks muddled.
Choosing an aspect ratio
The aspect ratio menu appears when Midjourney is the target, because Midjourney reads the --ar flag. For other tools, set the size in that tool’s own settings.
| Ratio | Shape | Common uses |
|---|---|---|
| 1:1 | Square | Profile pictures, feed posts, icons |
| 16:9 | Wide | YouTube thumbnails, slides, desktop wallpapers |
| 9:16 | Tall | Stories, Reels, Shorts, phone wallpapers |
| 4:3 and 3:2 | Classic photo | Blog images, prints |
| 2:3 | Portrait | Posters, book covers, Pinterest pins |
| 21:9 | Ultra-wide | Website banners, cinematic scenes |
Common prompt mistakes to avoid
- Stacking styles. “Watercolor, 3D render, pixel art” pulls the model three ways. Pick one medium.
- Writing “no” in the main prompt. “A beach with no people” can actually add people, because the model sees the word. Put unwanted items in the negative prompt for Stable Diffusion, or, for other models, describe the scene positively (“an empty beach”).
- Relying on quality keywords. “8k” and “masterpiece” help some Stable Diffusion checkpoints but do little in Midjourney or DALL·E. A clear subject matters far more.
- Expecting exact text in the image. Many models still misspell words inside images. Add lettering afterwards in an editor if it must be correct.
- Changing everything at once. When a result is close, change one setting at a time so you can see what made the difference.
More questions
What is a negative prompt?
It is a separate list of things you do not want in the image, such as blur, watermarks or extra fingers. Stable Diffusion interfaces have a dedicated field for it. Flux models usually don’t use a negative prompt, so describe what you want instead. Midjourney uses its own --no parameter instead, and DALL·E has no negative prompt field.
How long should a Midjourney prompt be?
Short to medium usually works best: one clear subject plus a handful of descriptive phrases. Very long prompts dilute each word’s influence, so cut anything that does not change the picture.
What does the “Surprise me” button do?
It fills in a random example subject and picks random options for style, lighting, framing, mood and palette. It is a quick way to explore combinations you might not have tried.
Why does the same prompt give different images each time?
Image generators start from random noise, so every run differs unless you fix the seed value in tools that support it. Use the prompt to set the direction, then generate several versions and pick the best.
Related tools: shrink your finished images for the web with compress image, change their dimensions with resize image, or find ready-made text prompts in the prompt library.
Further reading: Text-to-image model (Wikipedia).