Veo is Google's video generation model, and it turns a text prompt into a short clip. Writing the prompt as a small JSON object, with one field each for the shot, the subject, the motion, the light, the audio and the duration, gets you the shot you asked for more often than a paragraph does, because nothing gets buried in a sentence. This page gives you the template, a worked example and the failures each field prevents; the free guide adds 50+ JSON prompts and a mini-course.
What Veo makes
Veo makes one clip per prompt: a few seconds of motion from a written description, and depending on the version, from a starting image and with sound. The download covers Veo 3.
The clips worth making are single beats: a product turning in light, a character reacting, a drone push over a coastline, a loop for an ad. Stories are many clips cut together, each with its own prompt.
Why write a Veo prompt as JSON
Veo reads plain language; the JSON keys are for me.
A JSON prompt forces a decision for every field. A paragraph lets you skip the camera and the light without noticing; an empty "camera" field stares at you until you fill it. It makes iteration honest: change one value, run again, compare. And a folder of JSON prompts is a library you can search.
The template and one example
The keys are mine. Rename them if you like; the discipline is the point.
{
"shot": "framing, angle and lens feel",
"subject": "who or what, with two physical details",
"motion": "one action the subject performs, start to end",
"camera": "static, or one movement",
"light": "source, direction, time of day, mood",
"audio": "ambient sound, dialogue, music, or silence",
"duration": "seconds, and what the final frame shows",
"constraints": "aspect ratio, style, what to leave out"
}
Filled in:
{
"shot": "medium close-up, eye level, 50mm feel",
"subject": "a barista in a black apron, sleeves rolled, forearm tattoo",
"motion": "pours steamed milk into a cup, finishing a rosetta",
"camera": "static, shallow depth of field",
"light": "warm cafe light from the left, late afternoon",
"audio": "milk pouring, low cafe hum, no music",
"duration": "about 6 seconds, ending as the pour finishes",
"constraints": "16:9, photoreal, no text, no logo"
}
Paste it as the prompt, braces and all, or read the values out as sentences. Both work; the JSON is easier to edit.
The shot, motion and audio fields
Shot and camera split the job that "camera" usually carries alone. Shot is the framing (wide, medium, close-up, extreme close-up) and the angle. Camera is the movement: static, push-in, orbit, handheld follow. Keeping them apart stops you asking for a wide establishing shot and a slow push-in in one breath and getting neither.
Motion is one action with a start and an end. "Pours milk" is one action. "Pours milk, then hands the cup over" is two, and the second one is where the subject morphs.
Audio is the field people leave empty, and an empty audio field means the model picks for you. Name the ambient sound and say whether anyone speaks. Put "no music" if you plan to add your own. Dialogue works best as one short line in quotes, with who says it.
Common failures and the field that fixes them
| What you see |
The missing or wrong field |
The fix |
| The subject changes shape mid-clip |
Two actions in "motion" |
One action; make the second a new prompt |
| The camera cuts or drifts |
"camera" left empty |
Write "static" or name one movement |
| Flat, grey footage |
"light" left empty |
Name the source and time of day |
| Unwanted music or the wrong sound |
"audio" left empty |
Name the ambient sound and add "no music" |
| Garbled lettering on signs or packaging |
Text requested in "subject" or "constraints" |
Leave text out and add it in an editor |
| The clip stops mid-action |
No final frame in "duration" |
Write what the last frame shows |
When a clip comes back right, save the JSON next to it; next week you will have forgotten the wording.
What is inside the Veo guide
The free download has four parts.
- Veo Prompt Engineering Guide: the template above, field by field, with examples.
- 50+ Veo JSON Prompts: ready prompts in the format on this page.
- Veo Mini-Course: short lessons from a first clip to a repeatable session.
- Lifetime access and updates: every revision as Veo changes.
Drop your email in the form and it is yours.
The Gemini Omni guide covers the same fields written as plain text. For the assistant itself, start with the Gemini guide, and the Gemini prompts in the library cover the text side.
