Introduction
Text-to-image models learned the relationship between images and their descriptions from huge collections of picture–caption pairs. When you type a prompt, the model starts from random noise and refines it step by step into an image that matches your words (this family of models is called diffusion models).
The image prompt formula
| Part | Example |
|---|---|
| Subject | a young Bengali woman holding a clay cup of tea |
| Action / setting | at a roadside tea stall in north Kolkata, morning |
| Style / medium | photorealistic / watercolour / flat vector / 3D render |
| Composition | close-up, rule of thirds, space on the left for text |
| Lighting & colour | soft golden sunlight, warm tones |
| Format | square 1:1 / vertical 9:16 / landscape 16:9 |
Photorealistic close-up of a smiling young woman holding a clay kulhad of tea at a roadside tea stall in north Kolkata, early morning, soft golden light, shallow depth of field, warm tones, empty space on the left for text, 4:5 vertical
Popular tools
- ChatGPT image generation — conversational, good at following instructions and editing
- Google Gemini / Imagen — inside Gemini and Google tools
- Adobe Firefly — designed for commercial use, integrated with Photoshop/Express
- Canva AI — fastest route from image to finished design
- Midjourney — strong artistic quality (paid)
Rights and responsibility
- Read each tool's terms for commercial use.
- Do not generate real people without consent, brand logos, or copyrighted characters.
- Don't create misleading images of real events or products you don't sell.
- Label AI images where your platform or client requires it.