Use standard film vocabulary: name the shot size (close-up, medium, wide), the angle (low, eye-level, high), the lens effect (wide-angle, telephoto, shallow depth of field) and one camera movement (pan, dolly, tilt). Models trained on film and video descriptions respond best to common terms used once each, rather than a pile of exotic ones. This glossary follows the terms Google lists in its Veo prompt guide, with practical notes and Runway’s guidance on structure.
Shot size
Shot size controls how much of the subject and setting is visible.
| Term | What it frames | Use it for |
|---|---|---|
| Extreme close-up | A tiny detail | Texture, a logo edge, an eye |
| Close-up | Subject tightly | Product hero, emotion |
| Medium shot | Subject with some environment | Dialogue, product in use |
| Full or long shot | Whole subject and surroundings | Action, posture |
| Wide or establishing shot | Large scene | Place, scale, scene opening |
Google’s guide describes close-ups as framing the subject tightly and medium shots as balancing detail with some environmental context. Start with a clear shot size; ambiguity here causes more drift than any other term.
Camera angle
Angle sets viewpoint and mood. Google lists eye-level, low-angle, high-angle, bird’s-eye or top-down, worm’s-eye, Dutch or canted angle, over-the-shoulder and point-of-view shots. Low angles tend to make subjects feel imposing and high angles smaller, a convention from film that you can test on your own tool. The guide also warns that some advanced angles “are not officially supported”, so treat unusual angles as experiments and check outputs.
Camera movement
Name the movement and, if possible, the direction.
| Term | Meaning |
|---|---|
| Static shot | Camera does not move |
| Pan | Camera pivots left or right on a fixed point |
| Tilt | Camera pivots up or down |
| Dolly | Camera physically moves toward or away from the subject |
| Truck | Camera moves sideways |
| Pedestal | Camera moves vertically |
| Zoom | The lens changes focal length |
| Crane or aerial shot | Camera rises or flies over the scene |
| Handheld | Slightly shaky, documentary feel |
| Whip pan | Very fast pan, often as a transition |
| Arc shot | Camera circles the subject |
The difference between zoom and dolly is worth learning. Google’s guide notes that a zoom “is different from a dolly, as the camera itself doesn’t move”. A dolly changes the relationship between foreground and background, while a zoom only magnifies. If you want a sense of travelling through space, ask for a dolly.
Lens and optical effects
Google lists wide-angle lens, telephoto lens, shallow depth of field, deep depth of field, lens flare, rack focus, fisheye effect and the vertigo effect (dolly zoom). As a rule of thumb from conventional photography: wide-angle exaggerates depth and suits interiors and landscapes, telephoto compresses distance and flatters portraits, and shallow depth of field separates the subject from a soft background. Again, the guide flags that some advanced lenses are not officially supported.
Mapping terms to behaviour
Models are statistical, so these terms shift probabilities rather than operate like camera controls. Runway’s prompting guide gives three practical rules that apply broadly:
- Choose one primary movement per prompt. Runway warns that multiple camera directions can produce chaotic results, and recommends leading with the shot type.
- Describe what you want, not what to avoid. The guide says “No camera shake” may make the model focus on shake, and suggests “Smooth, stable camera movement” instead. For a locked frame, write “static shot, camera holds still”.
- Keep each prompt to one action. Runway suggests one action spanning roughly 5 to 10 seconds, with no “then” or “next”, and building complexity only after the basic move works.
Beyond that, results vary by model and version, so test your vocabulary on the tool you use and keep notes. “Close-up, shallow depth of field” is a common pairing worth trying first.
A simple prompt order
The structure in how to write prompts for cinematic AI video shots works well: shot size and angle first, then subject and action, then setting, lighting and style, with camera movement stated clearly near the start. For example (hypothetical): “Medium shot, low angle. A ceramic mug on a timber bench, steam rising. Slow dolly in. Shallow depth of field, soft morning light.”
Keep a shot library
Build a small internal glossary of phrases that reliably work on your tools and reuse them. This is also how you maintain visual continuity, as discussed in keeping characters and products consistent across AI shots, and how you plan short sequences like those in a cinematic 15 second ad.
What camera language cannot do
- It cannot guarantee a specific lens or real physics. Models imitate the look, not the optics.
- It cannot fix subject errors such as distorted hands or text.
- It cannot replace a shot list: decide what each shot is for before you prompt.
- Always review output for artefacts before it goes to a client, and apply your brand checks.
Where Radiance Impact fits
Radiance Impact’s Studio produces AI commercials and product films, brand films and short-form video, combining cinematic planning with AI tools. To discuss a project, get in touch.
Sources
- Video generation prompt guide, Google Cloud (Vertex AI documentation)
- AI video prompting guide, Runway
