Blog › AI Image Generator guide
How to Write Image Prompts That Actually Work
Everyone's first AI image is a disappointment: a muddy scene, the wrong subject, hands from a nightmare. The model was not being difficult — the prompt was being vague. AI image models are brilliant illustrators with zero imagination about what you meant. They draw exactly what the words most often mean. This guide gives you the formula that closes the gap, using our free AI Image Generator.
The four-part formula
Subject + Setting + Light/Mood + Style. Compare "a dog" (anything could happen) with "a golden retriever puppy (subject) sitting in a sunflower field (setting) at golden hour, warm and joyful (light/mood), photorealistic (style)". Four short decisions turn a dice roll into a direction.
- Subject first. The first words carry the most weight. Lead with what the picture is of.
- Setting grounds it. A place, a surface, a background — "on a wooden kitchen table" beats floating in void.
- Light is the cheapest upgrade. "Soft window light", "golden hour", "neon night", "overcast morning" — one light phrase transforms flat renders.
- Style is chosen for you in our generator — photo, illustration, 3D, anime or logo — so your prompt can spend its words on content.
Style cheat-sheet
- Photorealistic — product shots, places, people, anything that must look real. Add camera words: "close-up", "wide shot", "shallow depth of field".
- Illustration — blog headers, social posts, children's content. Flat, friendly, forgiving of odd details.
- 3D render — app icons, game-like scenes, product concepts. Clean surfaces, studio light.
- Anime — characters and scenes in Japanese animation style. Name hair, outfit and expression — anime prompts reward character detail.
- Logo — simple marks and icons. Keep the prompt minimal: "a fox head, geometric, orange and black". Words in logos still garble — for a name-based logo, use the AI Logo Maker which renders real text.
The five failures everyone hits — and the fixes
- Mangled hands and faces. Pull back: "wide shot" or choose illustration/3D, where the style hides it. Or crop the idea: an image of the object, not the person holding it.
- Gibberish text. Models cannot spell reliably. Design around it: signs and labels blank, add words later in any editor.
- The subject is tiny or missing. You buried it mid-prompt. Move it to the front and repeat its key adjective once.
- Everything is beige. Name two or three colours explicitly. Colour words are strong steering.
- It looks like a stock photo. Add one unusual, specific detail — "a bicycle with a basket of baguettes" — specificity is what stock photos never have.
Versions are free — use them
Every generation in our tool uses a fresh random seed, so "Make another version" genuinely re-composes the picture. Professionals generate four to eight versions and keep one. Think of the prompt as the brief and the versions as the drafts — that is how AI images stop being a lottery and start being a tool.
Start now with the free AI Image Generator — no sign-up, and every download is yours to keep.