Veo 3.1 is the most controllable video model you can point at a product — and that’s exactly why most Veo ads fail. Give it a vague brief and it reaches for what it does best: cinematic lighting, stabilized moves, color-graded gloss. Beautiful footage, terrible UGC. The entire value of user-generated-content ads is that they don’t read as ads; the moment your clip looks produced, you’ve paid for a commercial nobody trusts and the feed scrolls on.
This guide is the full anatomy of a Veo 3.1 prompt that makes the model shoot like a person with a phone instead of a production company. It’s the same system our free prompt generator applies automatically — read this to understand the levers, use the tool when you want them pulled for you.
Why Veo needs structure, not vibes
Veo 3.1 responds dramatically better to labeled sections than to a paragraph of adjectives. Where Seedance rewards a shot list, Veo parses structure: give it discrete Subject, Action, Scene, Camera, Lighting, Style, Dialogue, Audio, and Negative prompt lines and it honors each one with surprising obedience. That obedience is your weapon: UGC authenticity isn’t a mood you request, it’s a checklist you specify.
The order matters less than the completeness. Every section you leave out is a decision you’ve delegated back to a model whose taste runs cinematic. Leave out lighting and you’ll get three-point studio glow. Leave out audio and you’ll get a music bed. Leave out the negative prompt and you’ll get both.
The eight decisions that make it UGC
1. Subject: a person, not a presenter
Cast a believable owner of the product, described with texture: “a woman in her late 20s with tired-but-happy morning energy, hair in a claw clip” beats “an attractive young woman”. Age bands, energy, and one imperfect human detail (a stretched-out hoodie, yesterday’s nail polish) push Veo away from stock-footage casting. Match the persona to the buyer — a skeptical man in his 50s sells a back-pain product harder than any model ever will.
2. Action: hands and product, in that order
Write the physical beats explicitly: she “holds the serum up to the lens so the label catches the light, pumps one drop, pats it into her cheek”. Concrete micro-actions do two jobs — they give Veo motion to render instead of a static talking head, and they put the product in frame naturally, the way real reviewers handle things.
3. Scene: lived-in, never staged
The give-away of fake UGC is a clean background. Ask for the mess: “a real home kitchen, everyday clutter softly out of focus — mug tree, paper-towel roll, kids’ drawing on the fridge”. One line of set dressing buys more authenticity than any style keyword.
4. Camera: one shot, one movement
This is the most common Veo mistake: stacking moves (“dolly in, then orbit, then rack focus”) produces the cinematic look you’re trying to kill. UGC is one handheld framing with micro-instability: “handheld selfie medium close-up, subtle natural sway, one quick reframe mid-shot”. If the phone would be propped on a shelf, say that instead — static with tiny micro-shake reads perfectly real.
5. Lighting: name a household source
Never say “natural lighting” — Veo hears “beautiful lighting”. Name the actual source and its flaw: “window daylight from the left, slightly uneven exposure, no fill”. Bedside lamps, kitchen overheads, car daylight through a windshield — imperfection is the point. If the exposure is a little wrong, the clip is a little true.
6. Dialogue: write the exact words, in quotes
Veo 3.1 generates speech, so script it fully: hook inside the first two seconds, 20–35 words total for an 8–10 second clip, contractions and filler energy included. “Okay so — ice from yesterday? Still ice. I don’t make the rules.” Add a delivery note (“casual, amused, like telling a friend”) and specify direct-to-camera eye contact. What you must never do is describe the topic of speech (“she talks about the bottle’s insulation”) — that’s how you get announcer-voice paraphrase.
7. Audio: phone-mic honesty
Ask for the room: “kitchen room tone, ice rattling near the mic, phone-mic voice quality, no music”. A UGC clip with a music bed is an ad with a costume on. If you want sound design, let it be diegetic — the product’s own clicks, pours, and zips sell harder than any track.
8. Negative prompt: your anti-ad insurance
End every Veo UGC prompt with explicit negatives: studio microphone, cinematic color grading, tripod stability, polished ad look, announcer voice, teleprompter delivery, ring-light glamour, on-screen text, watermark. This single line does more to protect authenticity than everything above it combined, because it fences off Veo’s defaults by name.
A complete example
Here’s the checklist assembled, for an insulated bottle testimonial:
Subject: A woman in her late 20s with tired-but-happy morning energy, holding a matte-black insulated bottle. Action: She talks into her phone's front camera, taps the bottle lid twice, takes a sip, and grins. Scene: A real home kitchen, everyday clutter softly out of focus. Camera: handheld selfie medium close-up, subtle sway. Lighting: window daylight from the left, slightly uneven. Style: authentic smartphone UGC, natural skin texture, no retouching. Dialogue: "Okay so — ice from yesterday? Still ice. I don't make the rules, it just works." Audio: kitchen room tone, ice rattle close to the mic, phone-mic voice. Negative prompt: studio microphone, cinematic color grading, polished ad look, teleprompter delivery.
Notice what’s absent: no resolution requests, no “8k”, no “masterpiece”. Quality keywords drag Veo back toward gloss. Specify reality; the model supplies the fidelity on its own.
Structure the ad, not just the shot
A technically authentic clip still fails without ad architecture. Keep the classic UGC skeleton — Hook (0–2s) → Body → CTA — inside your dialogue. The hook is a question, a bold claim, or a named frustration; the body is one believable, specific, non-medical proof point; the CTA is soft (“honestly just try it”) because hard sells break the friend-recommendation illusion. Our Prompt Library has this structure pre-built for twelve product categories if you’d rather start from a working template.
The five mistakes that reintroduce the ad look
- Stacked camera moves. One movement, maximum. Two reads as a crew.
- “Professional lighting” anywhere in the prompt. Name a window or a lamp instead.
- Describing dialogue instead of scripting it. Paraphrase in, announcer out.
- Music beds. Room tone plus product sounds, nothing else.
- Skipping the negative prompt. Veo’s defaults are cinematic; unfence them and they return.
Text prompt or JSON?
Veo 3.1 accepts both, and they’re not rivals — they’re stages. The labeled-text version is where you think: easy to read, easy to edit one line at a time. The nested JSON version (video → description, camera, lighting, dialogue, audio, negative_prompt) is where you standardize: once a structure works for your product category, JSON makes every field explicit, diffable, and safe to hand to a teammate or a pipeline without the model re-interpreting loose prose. A sane workflow: iterate in text until the clip lands, then keep the JSON twin as your reusable master. PromptShot outputs both shapes for every generation precisely so the handoff costs nothing.
Iterate like an editor, not a gambler
When a generation misses, change one section and re-run — swap the hook line, or the setting, and keep everything else frozen. Re-rolling the whole prompt teaches you nothing about which lever moved the result. This is also why regenerating with a deliberately different angle (new hook framing, same authenticity rules) out-performs re-rolling identical text: variation where it matters, consistency where it counts.
Or skip the manual work entirely: upload a product photo to PromptShot and it writes the full Veo 3.1 prompt — sections, script, negatives — plus matching Seedance 2.5 and Kling versions in the same click. Five free generations a day, no account.