AI Image and Video Prompt Formula: 7 Parts of a Good Prompt
A good AI image or video prompt names seven things: the subject, what it is doing, the setting, the camera framing or movement, the lighting, the visual style, and any constraints. Keep one idea per prompt, describe what you want to see rather than what you do not, and change one variable at a time when you iterate.
The 7-part prompt formula
| Part | What to specify | Example |
|---|---|---|
| 1. Subject | Who or what, with specific details | a young woman in a mustard jacket |
| 2. Action | One clear moment or movement | laughing while reading her phone |
| 3. Setting | Where and when | rooftop cafe at golden hour |
| 4. Camera | Shot size, angle, lens feel, movement | medium close-up, eye level, slow push-in |
| 5. Lighting | Quality and direction of light | soft window light from the left |
| 6. Style | Look, medium and colour palette | natural documentary look, muted warm colours |
| 7. Constraints | What to keep or avoid | single subject, no text on screen |
As a template: [Subject] [action], [setting], [camera], [lighting], [style]. [Constraints].
Before and after examples
| Weak prompt | Stronger prompt |
|---|---|
| a nice video of a coffee shop | A barista pouring latte art into a white cup, small wooden cafe at sunrise, close-up from the side, shallow depth of field, soft warm window light, natural documentary style, no text. |
| futuristic city | Wide aerial shot of a dense futuristic city at night, rain-slick streets reflecting neon signs, slow forward camera movement, cinematic teal and orange palette. |
| a product ad for a bottle | A matte black water bottle on a wet slate surface with water droplets, low-angle close-up, single soft key light from the right, minimal dark background, premium commercial look. |
Results vary by model, so treat these as starting points rather than guaranteed outputs.
Image prompts vs video prompts
An image prompt describes a frozen frame: subject, composition, light and style.
A video prompt adds motion. Describe what moves and how the camera moves, and mention the speed ("slow", "steady"). When you animate an existing image, do not re-describe it. Describe only the motion, one action per clip.
A good workflow is to perfect the still first, then animate it. The 7-step AI video workflow shows where prompts fit in the full process.
A simple iteration method
- Generate three or four variations of the same prompt.
- Pick the closest one and note what is wrong with it.
- Change one variable, such as camera or lighting, and generate again.
- Save the winning prompt as a template.
- Keep one fixed style sentence across every shot of a video so the shots match.
Common prompt mistakes
- Walls of text. Specific beats long. Extra adjectives often add noise.
- Contradictory styles. "Photorealistic anime watercolour" gives the model nothing clear to follow.
- Vague praise words. "Beautiful" and "amazing" do not describe anything.
- Readable text inside images. This is often unreliable. Add text in your editor instead.
- Several actions in one clip. Split them into separate shots.
- Only saying what you do not want. Describe what you do want first.
Build a personal prompt library
Keep a simple document with the prompt, the tool and model used, the result, and a note on what to change. Group prompts by shot type, such as close-up, wide establishing shot, product shot and talking head. Over time this becomes your fastest way to produce consistent work.
Frequently asked questions
Do longer prompts work better?
Not necessarily. Specific beats long. Cover the seven parts concisely, since extra adjectives often add noise rather than control.
Can ChatGPT write prompts for image and video tools?
Yes. Describe your shot, ask it to write a prompt using the seven parts, then edit the result. You stay in charge of the final wording and check it against your tool's behaviour.
Why do my characters look different in every shot?
Most models do not remember earlier generations. Reuse the same detailed character description and style sentence, and use reference-image or character features where your tool offers them.
Should I use negative prompts?
Some tools support them and others do not. Where supported, use them briefly for specific problems. The main prompt should still describe what you want to see.