The basics of prompting
This chapter introduces the core principles of prompting the part that stays relevant even as tools and models evolve.
1. What is prompting?
Prompting is the act of communicating with a generative AI system the way you express what you want it to create.
A prompt is your creative direction in language form.
Think of it as briefing your most literal, tireless assistant:
it will do exactly what you say but only if you say it clearly.
A well-crafted prompt:
- provides context and purpose
- establishes style and tone
- defines constraints (e.g. duration, lighting, format)
- allows room for interpretation when needed
2. The anatomy of a strong prompt
Every strong prompt combines five ingredients:
| Element | Description | Example (text) | Example (photo/video) |
|---|---|---|---|
| Goal | What should be created | “Write a short email invitation for a networking event.” | “Create a cinematic portrait of a dancer in golden light.” |
| Context | Why, for whom, in what setting | “…for creative professionals attending a media innovation workshop.” | “The dancer is rehearsing alone in an empty theatre at sunrise. |
| Style | Visual or emotional tone | “Friendly, professional, and slightly playful.” | “Dreamy, analog film look, soft focus, golden-hour tones.” |
| Structure | Technical details | Include a title, event date, and RSVP link.” | “4K, 16:9, 24fps, shallow depth of field.” |
| Iteration (optional) | Request multiple versions or refinements | Generate 3 alternative subject lines and closing paragraphs. | “Generate three camera angles with different lighting directions.” |
3. Prompt types: from simple to structured
| Type | Description | Example |
|---|---|---|
| 🟢 Natural language/Text prompt | Free, descriptive instructions | “A woman walks through a rainy Tokyo street, cinematic lighting.” |
| 🟣 Structured prompts/JSON prompts | Information formatted in JSON or parameters, useful for automation or APIs. | { "scene": "forest", "lighting": "dawn", "duration": "6s" } |
Learn more about structured prompts in the separate chapter JSON Prompting
Advanced prompting techniques
Once you understand the two main prompt types, you can combine them with advanced techniques to gain more creative control:
Few-shot prompting: give examples before asking for output
What it is:
You provide one or more examples before asking for a new output. The AI learns from the examples to understand style, tone, or structure.
When to use it:
When you want consistency across outputs. For example, a series of social videos, a recurring character, or a visual identity.
Where it works:
Mainly in text-based assistants (ChatGPT, Claude, Gemini). Useful for developing scripts, shot descriptions, or visual briefs that later feed into image/video tools like Runway, VEO3, or Midjourney.
Example:
Here are two examples of short video voice-overs we liked:
Example 1: “Ever wished your workspace felt alive? Bring your ideas to motion — fast, fun, frictionless.”
Example 2: “From sketches to screen — your vision, rendered in seconds.”
Now write a new 10-second script for a video introducing a fictional AI tool called Cinemagic.
Keep the same rhythm, sentence length, and energetic tone.
Persona prompting: define the AI’s role (e.g. “You are a cinematographer”)
What it is:
You define a role or persona for the AI to act as.
This shifts how it interprets tone, vocabulary, and creative decisions.
When to use it:
When you want the AI to think like a specialist (director, colorist, writer) or to emulate a creative style.
Example:
You are a cinematographer specializing in natural light and handheld camera work.
Suggest three camera setups for a short film scene shot in an old factory.
Each setup should include lens choice, lighting direction, and camera movement.
Prompt chaining: use output from one step as input for the next
What it is:
You use the output of one prompt as the input for another, creating a workflow of dependent steps. Common in multimodal pipelines (e.g. ChatGPT → Midjourney → Runway).
When to use it:
When building complex creative outputs that evolve through stages: concept → design → animation → edit.
Example chain:
Prompt 1 (text):
Write a short narrative concept for a 20-second ad about renewable energy. Include characters and visual motifs.
Prompt 2 (image):
Using the concept from step 1, generate a storyboard of three key shots in Midjourney.
Prompt 3 (video):
Animate the first storyboard frame into a 5-second clip using Runway. Keep lighting and color palette consistent with the previous image.
Meta-prompting: ask AI to improve your prompt
What it is:
You ask the AI to analyze and improve your prompt before using it. This turns the model into a prompt editor or coach.
When to use it:
When you want to polish your prompt for clarity, focus, or cinematic quality.
Example:
Here’s my current prompt:
“Create a short film scene with two people arguing in a café.”
Please rewrite this prompt to make it more cinematic, visually descriptive, and emotionally engaging.
Add camera directions and lighting style.
Constraints: define what to include/exclude clearly
What it is:
Explicitly defining what should or should not be included in the output. Constraints focus the AI and prevent unwanted artifacts or style drift.
When to use it:
When working with video or image generation where composition, timing, or style must stay consistent.
Example:
Create a 10-second cinematic shot of a skateboarder in slow motion.
Constraints:
– No text or logos.
– 24 fps, 1080p, 16:9 aspect ratio.
– Daylight only, no night scenes.
– Keep the camera at waist height, moving parallel to the subject.
4. Tips and tricks
These universal best practices will outlast any single AI tool:
Be specific – precise language = precise results
Precise language = precise results.
Avoid vague instructions like “a cool video”. Instead, describe what, how, and why.
Why:
AI models respond better to concrete visual and emotional cues.
Example:
“A nice car driving fast.”
“A red sports car racing down a coastal road at sunset, filmed from a drone at 24 fps with lens flare and warm tones.”
Add context – describe the goal and audience.
Describe the goal, audience, and medium so the AI understands tone and purpose.
Why:
The same visual idea looks completely different in a YouTube ad vs. a museum installation. Context drives framing and energy.
Example:
“Create a 15-second vertical video for TikTok promoting a sustainable fashion brand. The audience is 18–25-year-old creators; keep it upbeat, colorful, and text-driven.”
Use style references – films, artists, genres.
Ground your request in recognisable visual, cinematic, or artistic styles.
Why:
References anchor the AI’s aesthetic choices and reduce randomness.
Example:
“Render a city skyline in the visual style of Blade Runner 2049,
using Roger Deakins-style lighting — moody, contrasty, orange-blue palette.”
Give constraints – fps, length, framing, tone.
Define technical or stylistic boundaries : duration, resolution, camera angle, tone.
Why:
Constraints simulate real-world production limits and make outputs usable in actual workflows.
Example:
“Generate a 6-second slow-motion shot at 24 fps, 16:9 ratio, daylight only.
Keep the subject centered, no logos or text.”
Iterate – test, review, and refine your prompt.
Don’t expect the first result to be perfect, treat AI like a collaborator.
Why:
Iteration lets you discover what the model truly understands and improves creative control.
Example:
Start: “A dancer performing under neon light.”
Refine: “Add camera movement, more depth, and cinematic color contrast.”
Final: “Handheld shot, 35 mm film look, shallow depth of field, rhythmic editing.”
Ask for variations – explore alternatives.
Request multiple options with controlled diversity.
Why:
Variations encourage exploration without starting from scratch — like having multiple drafts on set.
Example:
“Generate three versions of this concept:
– one romantic,
– one futuristic,
– one documentary-style.”
Use positive phrasing – say what you want, not what you don’t want.
State what you want, not what you don’t want.
AI handles negative statements poorly.
Why:
Positive phrasing keeps the model focused on what to produce, not what to avoid.
Example:
“Don’t make the lighting dark.”
“Use bright, diffused daylight with soft shadows.”
Reuse good prompts – build a personal prompt library.
Build a personal prompt library of effective structures, tones, and keywords. Why:
Consistent prompts = consistent visual identity across projects or brands
Let AI help you – ask it to optimize your prompt.
Ask the model itself to improve your prompt or generate better alternatives.
Why:
Meta-prompting helps you learn how the AI interprets creative direction.
Example:
“Rewrite my prompt to sound more cinematic and professional:
‘A person standing in the rain looking at city lights.’”
Stay in control – AI is your co-creator, not your replacement.
AI is a co-creator, not a replacement.
Use it to accelerate iteration, not to delegate vision.
Use AI to generate concept art or first cuts, but make final aesthetic and narrative decisions yourself.
Why:
Creativity remains human-driven, AI extends, not replaces, your authorship.