Most disappointing AI images are not a model problem, they are a description problem. "A nice photo of a coffee cup" leaves the model to invent the light, the lens, the surface and the mood, and it will invent them the same way it invents them for everyone else. This builder makes you choose those things, then writes them in the order image models actually read.
You pick a subject and a light, a lens, a medium and a palette. The generator writes one tight paragraph plus the settings that travel outside the prompt, and it explains which word is doing the work so the next prompt you write on your own gets better. Nothing is uploaded, no account is created, and no image is generated here — this page writes the text, your own tool makes the picture.
Build a prompt
Nothing here talks to a server: the wording is assembled in your browser from the choices below, so it costs nothing and works with an ad blocker on.
One tight paragraph for an image model, plus the settings to send with it.
One noun phrase, not a sentence. Name the thing and the one detail that matters.
Extra craft words
Optional. These are the words that separate a stock-looking result from a deliberate one.
Image prompt
a single ceramic coffee cup, a seamless studio backdrop, soft directional window light, 35mm documentary framing, photograph, minimal black and gold palette, calm mood.
One paragraph, ordered subject-first. Image models distort when prompts run long — under about 200 tokens is the range where this stays reliable.
Send it with these settings
Aspect ratio: 1:1 — square, for profile images and grid posts. Model family: any text-to-image model that accepts a free-text description.
Settings travel outside the prompt text for almost every image model.
Things to add yourself
Instead of "no blur" — write "tack sharp focus". Instead of "no people" — write "uninhabited landscape". Instead of "not dark" — write "bright, evenly lit". Then generate four variations and change exactly one word between them, so you learn which word did what.
Most image models ignore a negative prompt and quietly do the opposite. State what you want, not what you do not want.
No account, no API key, no generation. This page only writes the text — you paste it into whichever tool you already pay for. Nothing you type here is uploaded, and this site is not affiliated with any generation service.
How to use
- Describe your subject as a noun phrase, not a sentence: "a single ceramic coffee cup", not "I want a picture of a cup".
- Choose a setting and a light. Light changes more about the result than any adjective you can add.
- Pick a lens or framing and a medium, so the model knows whether it is shooting a photograph or painting an illustration.
- Add a palette and mood to lock the colour range and the feeling.
- Copy the prompt, send it with the aspect ratio the tool reports, then change exactly one word between runs so you learn what each word does.
Frequently asked questions
Is this generator free, and does it ask for an API key?
It is free and there is no key or account. The wording is assembled in your browser from a fixed vocabulary, so there is nothing to bill and nothing to rate-limit.
Does the tool create the image?
No. It writes the prompt. You paste that text into whichever image tool you already use. Keeping the two steps apart is what lets the generator stay free.
Why is the output longer than the prompts I normally write?
Because naming light, lens, medium and palette replaces guesswork with instructions. That said, longer stops helping past roughly 200 tokens — image models start dropping parts of a long prompt rather than obeying all of it. Trim the words that matter least to you.
Can I ask for a style that copies a specific artist?
The vocabulary deliberately avoids artist, band and brand names, and prompts built from them are often rejected outright as containing protected references. Describing the qualities you are after — medium, era, light, texture, palette — gets you closer anyway, without betting your project on a name.
The result still is not what I pictured. What now?
Change one variable per run, in this order of impact: subject wording, then lighting, then lens, then medium, then palette. Changing three things at once gives you a different picture but no information about why.
Should I use a negative prompt?
Usually no. Most image models either ignore a negative prompt or handle it weakly, and "no blur" has a way of producing blur. State the positive version instead — "tack sharp focus", "uninhabited landscape" — which the builder's tips section spells out.