Describe it, see it

Text-to-image tools are everywhere and most of them ask for the same three things before they will show you a picture: an account, a card, and patience. This generator asks for a sentence. Type what you want to see, pick the look and the shape, and a real generated image comes back in a few seconds — a photograph of a harbour at dawn, an illustration for a blog header, a poster texture, a background for a slide. Nothing is traced from a stock library and nothing is drawn from a template: the picture is generated from your words.

What "look" actually changes

The look is not a filter laid over the result; it is part of the instruction the model receives. Photograph adds the language of a camera — natural light, sharp focus, 35mm, realistic colour — and that is what makes a prompt about a street come back looking like a street rather than a painting of one. Illustration trades the camera for shapes and flat colour, which is what a blog header needs. Cinematic adds side light and depth of field, the reason film stills feel like film stills. A 3D render adds matte materials and studio light; flat vector and poster drop shading entirely; texture turns any subject into a pattern you can lay under white text.

Because the look carries the technical words, your own prompt can stay human. "A quiet fishing village at sunrise, wooden boats, mist over the water" is enough. You do not need "8k, ultra detailed, masterpiece" — that vocabulary stopped helping years ago and mostly makes a picture look busy.

Prompts that work, and prompts that do not

Describe what is in the frame before you describe how it should feel. Subject, place, light, and one detail is the shape that lands most often: a fox in snow at dawn, a bicycle leaning on a blue wall in afternoon sun, a market stall lit by one bulb. Light matters more than any adjective, because it decides contrast, colour and mood all at once.

What fails is abstraction. "Success", "innovation", "the future of work" give the model nothing visible, so it reaches for the nearest cliché — a glowing lightbulb, a handshake, a rocket. If you need a concept picture, translate it into a scene first: instead of "teamwork", write "four people carrying one long table up a narrow staircase". The result will be specific, and specific is what people stop scrolling for.

Sizes, and where each one is used

Square (1:1) is the safe default: avatars, album covers, product tiles, feed posts that get cropped from every side. Portrait (4:5) is the tallest shape Instagram shows in the feed and the closest thing to a printed flyer. Wide (16:9) is for slide decks, video covers, blog headers and anything that will sit behind a title. Phone (9:16) is stories, reels covers and lock screens. Pick the shape before you generate: a wide picture cropped to a square loses the composition that made it work.

Variations, seeds and getting the same picture twice

Every picture carries a seed — the number that decides the random choices inside the model. Keep the prompt and the seed and you get the same image back, which is how you build a set that matches: one cover per episode, same look, same palette. Press "New variation" and only the seed changes, so the direction stays and the composition moves. When you find a prompt that works, save it somewhere; the second half of a good result is usually a small edit to a prompt you already had.

How fast, and why there is a daily limit

Generation runs on a server, not in your browser, so the first picture usually takes a few seconds and a repeat of the same prompt is instant — the file is already saved. Every visitor gets a set number of pictures per day. The limit exists because each image costs real computing time; it is counted separately from the writing tools, and it resets every day.

What you can do with the file

The download is a clean JPEG at the size you chose, written without a watermark, and the file is yours: use it in a blog post, a cover, a slide, a shop listing, a print. What you should not do is treat a generated picture as a photograph of something real. It is an illustration of a description — do not present it as a photo of a product you sell, a place you visited or a person you interviewed, and do not prompt for a recognisable brand, logo or living person.

When a picture is the wrong tool

If you need a diagram, a chart, a screenshot or a logo with readable text, a generator will disappoint you: models still cannot draw letters reliably. Use a chart tool, a screenshot, or the logo studio on this site, which sets the name in real type instead of asking a model to spell it. Generated images are for atmosphere, illustration and background — the places where a made-up scene beats an accurate one.