How to Write Prompts for AI Image Generators (with 40 Examples)
By the NextG Design team · Published 19 October 2026 · 3 min read
Quick answer: A good AI image prompt names the subject first, then adds the style or medium, the lighting, and the camera or composition — for example “a red fox sitting in autumn leaves, wildlife photo, soft morning light, close-up, shallow depth of field”. Keep it under about 60 words, use concrete visual nouns and adjectives rather than instructions, and generate several variations because each seed gives a different result.
The prompt formula
Most strong prompts follow the same order, because image models give more weight to the words that come first:
- Subject: who or what — “an old fisherman”, “a glass perfume bottle”.
- Details: what makes it specific — “with a grey beard and yellow raincoat”, “on a silk cloth”.
- Style or medium: “documentary photo”, “watercolor painting”, “3D clay render”.
- Lighting: “golden hour”, “soft window light”, “dramatic rim light”.
- Camera & composition: “close-up portrait”, “wide angle”, “seen from above”, “shallow depth of field”.
Examples by category
Photos
- “a cozy cabin in a snowy pine forest at dusk, warm light in the windows, photo”
- “a busy night market in Taipei, neon signs, light rain, street photography”
- “a misty lake at sunrise with a wooden rowing boat, calm water, reflections”
- “a vintage red car on a coastal road, 1970s film photo, warm tones”
Products and food
- “a perfume bottle on white silk, studio product photo, soft shadows”
- “a bowl of ramen with a soft-boiled egg, overhead food photography, dark wood table”
- “a pair of white sneakers floating, pastel background, minimalist product shot”
- “a ceramic coffee mug on a marble counter, morning light, lifestyle photo”
Art and illustration
- “a lighthouse on a cliff, watercolor painting, loose brush strokes”
- “a cute robot reading a book under a tree, children's storybook illustration”
- “a fantasy castle above the clouds, digital painting, epic scale”
- “a cat in a spacesuit, retro 1950s poster art”
Backgrounds for the AI Background Generator
- “a modern minimalist studio, soft grey backdrop”
- “a sunny beach with palm trees, shallow depth of field”
- “a cozy library with wooden bookshelves, warm light”
- “a field of lavender at sunset”
Words that change the look
| You want | Add words like |
|---|---|
| Realism | photo, 35mm, natural light, detailed, documentary |
| Drama | dramatic lighting, rim light, low angle, stormy |
| Calm | soft light, pastel, misty, minimal, overcast |
| Depth | shallow depth of field, bokeh, foreground leaves |
| Illustration | watercolor, gouache, ink, cel shading, flat vector |
| Era | 1970s film, Victorian, retro-futurist, art deco |
Common mistakes
- Writing instructions. “Please make a picture where…” wastes words. Describe the picture itself.
- Too many subjects. Two or three things in a scene work; ten don't. Build complex scenes in layers with Generative Fill.
- Expecting legible text. Small image models are poor at lettering. Add text afterwards with the Type tool.
- Negative words. “No people” can make people appear, because the model sees the word “people”. Describe what you want instead — “an empty beach”.
- Real people and brands. Avoid prompts that imitate real, identifiable people or trademarks.
Generating images on your own device
The NextG Design AI Image Generator runs SD-Turbo on your graphics card with WebGPU, so prompts and results stay private. Choose a shape (square, portrait, landscape, story), a quality level, and optionally sharpen the result to around 2048 pixels with the AI upscaler. Results can open as a new image or be added as a layer to a photo you're editing.
Frequently asked questions
How long should an AI image prompt be?
Usually 10–60 words. Put the most important subject first; after about 75 tokens the model ignores the rest.
Why do I get different images from the same prompt?
Each generation starts from random noise (a seed). Generating several variations is the normal way to find the best result.
Can the AI write text in images?
Not reliably. Generate the image without text, then add lettering with the editor's Type tool for clean, editable text.
Is the AI Image Generator free and private?
Yes. It runs on your device, so there are no credits and your prompts aren't sent anywhere. The model downloads once (about 2.6 GB).
Can I sell images I generate?
Check the model's licence. SD-Turbo is under the Stability AI Community License, which allows commercial use for individuals and organisations under US$1 million annual revenue, with attribution; larger organisations need a licence from Stability AI.
Try it free: the AI Image Generator runs entirely in your browser — no upload, no account, no watermark.