Step 1: Start with the six-part formula
Write your prompt in this order. You do not need every element every time, but the more you specify, the less the model has to guess.
| Order | Element | Ask yourself |
|---|---|---|
| 1 | Subject | Who or what is on screen? |
| 2 | Action | What is happening? |
| 3 | Scene | Where and when? |
| 4 | Camera | Shot size and movement? |
| 5 | Lighting | Direction, time, mood? |
| 6 | Style | Look, finish, audio? |
Step 2: Compare weak vs strong prompts
The same idea, written two ways. The strong version simply answers more of the formula.
Weak
a dog running
Strong
A golden retriever sprints along a beach at sunset, low tracking shot beside it, 50mm lens, warm golden-hour backlight, sand spray in slow motion, joyful cinematic tone, gentle waves and barking audio.
Weak
a coffee cup on a table
Strong
A steaming ceramic coffee cup on a rustic wooden table by a rainy window, slow macro push-in, soft diffused morning light, shallow depth of field, cozy warm grade, ambient rain audio.
Weak
a city at night
Strong
A rain-slicked downtown street glowing with neon at night, slow elevated dolly forward, wide 24mm lens, reflections on wet pavement, cool cyberpunk grade, distant traffic and rain audio.
Step 3: Avoid the common mistakes
Stacking conflicting moves. "Dolly in while panning while orbiting" gives the model an impossible average. Pick one move.
Adjective soup. Ten mood words dilute each other. Three precise cues beat ten vague ones.
Ignoring the camera. No shot type or move means a default static frame. Always name one.
Forgetting audio. Veo 3 scores your clip if you ask. A missing audio cue leaves it to chance.
Burning credits in Quality mode while testing. Iterate in Fast mode first, then finalize.
Step 4: Iterate like a director
Treat your first render as a draft. Change one variable at a time — swap the lens, then the light, then the grade — so you learn what each word does. Because veo3gen bills pay-as-you-go at 1 credit per $0.01, a handful of Fast-mode tests costs very little before you commit an 8-second 1080p final. See the model comparison to choose between Veo 3 and Veo 3.1 variants.
Frequently asked questions
How long should a Veo 3 prompt be?
Aim for one to three rich sentences. You want enough detail to cover subject, action, scene, camera, lighting and style, but not a wall of contradictory adjectives. If a phrase does not change the image, cut it.
Should I describe the audio in the prompt?
Yes. Veo 3 generates native synchronized audio, so a short cue like "ambient city traffic and a distant siren" or "soft acoustic guitar" tells the model how to score the scene. Keep it to one clause.
Why does my prompt produce something different each time?
Generation is probabilistic, so two runs of the same prompt will differ. Lock down what matters by being specific, then iterate cheaply in Fast mode before committing a final render in Quality mode.
Can I reuse a character across clips?
Describe the character consistently — same wardrobe, hair, and distinguishing features in every prompt. Veo 3.1 improves consistency further, and starting from a reference image (image-to-video) gives the tightest match.
What resolution and length can I get?
veo3gen supports 4, 6 and 8-second clips, with up to 1080p on non-Lite models in 16:9 and 720p otherwise. Pricing is pay-as-you-go credits at 1 credit per $0.01.
Write one, render one
Put the formula to work in the generator and watch your prompt become video.