Veo Prompt Theory Prompts

Prompt Theory is a viral AI-video format in which AI-generated characters realise — or refuse to accept — that they are AI-generated. A person behaves completely normally, then stops, looks straight into the lens, and says one short line that breaks the frame. It spread on Veo 3 because Veo’s native audio makes the direct-to-camera delivery land, and because the format needs no visual effects at all: the whole thing is written into the prompt.

The prompts below are starting structures for that format. Copy one, change the bold details, and keep the shape.

The structure behind the format

Every working Prompt Theory clip has the same five beats, and skipping any of them is why most attempts fail:

  1. An ordinary setting. A street, a kitchen, a train. The more mundane the better — the break needs somewhere to fall from.
  2. Normal behaviour first. The character is doing something unremarkable and doing it convincingly.
  3. The stop. A named moment where the action halts. This is the beat people forget to write.
  4. One short line to camera. Under about ten words. Longer lines compress the delivery and lip-sync drifts.
  5. A defined reaction after. Returns to the task, holds the stare, or waits. If you do not specify this, Veo holds the look to camera for the whole clip.

What to avoid

Do not ask for cinematic grading, dramatic music, or glitch effects. The format works because it looks like ordinary footage that went wrong, and every layer of style you add moves it toward looking like a deliberate music video. no background music, no on-screen captions, locked-off camera belongs in almost every prompt on this page.

Do not write the emotion into the quotation marks either. says, in a quiet unsettled voice: "Who is writing this?" works; says "Who is writing this?" nervously frequently ends up with the word spoken aloud.

Where these came from

The format was popularised by AI creators on TikTok and X through 2025–2026 using Veo 3, most visibly a widely-shared series in which characters across different settings react to being told they are AI-generated responses to prompts. The prompts here are our own reconstructions of the structure rather than copies of any specific creator’s prompt.

Prompt deck

Copy a format, check the evidence, then customize it.

12 prompts 0 evidenced 0 community 0 owner-tested

Veo / Prompt Theory / self-aware AI characters

The original format — street interview, sudden awareness

Open in builder
Prompt
Handheld documentary shot on a **busy city sidewalk**. A **man in his thirties wearing a grey jacket** is mid-sentence answering an off-camera question when he stops, looks directly into the lens, and says in a quiet, unsettled voice: "Wait. Who is writing this?" He holds eye contact with the camera. Overcast daylight, natural handheld motion, photoreal. Ambient street noise, distant traffic, no music. No on-screen captions.

TweakThe whole format lives in the turn — normal behaviour, then the break to camera. Keep the line under ten words so the delivery lands inside the clip.

Veo / Prompt Theory / self-aware AI characters

Comedy club — the bit stops being a bit

Open in builder
Prompt
Medium shot from the back of a **small comedy club**, warm stage light on a **woman in her forties holding a microphone**. She delivers a punchline, the room laughs, and her smile fades. She looks past the audience into the camera and says flatly: "None of you are real either." Beat of silence. Photoreal, slight grain, stage-lit. Room tone, scattered laughter that cuts out, no music.

TweakThe laughter cutting out is doing the work. Name the audio change explicitly or Veo will hold the laugh track under the line.

Veo / Prompt Theory / self-aware AI characters

Restaurant kitchen — mid-task realisation

Open in builder
Prompt
Close-up in a **busy restaurant kitchen**, steam and motion behind. A **line cook in a white jacket** is plating a dish with fast, practised hands. He slows, sets down the tweezers, and says to no one: "I don't remember learning this." He looks at his own hands. Harsh overhead kitchen light, photoreal, shallow depth of field. Kitchen clatter, ticket printer, extractor hum, no music.

TweakLooking at their own hands is the strongest visual beat in this format — it reads as a body the character does not trust.

Veo / Prompt Theory / self-aware AI characters

Refusal — the character argues back

Open in builder
Prompt
Medium close-up. A **young woman in a denim jacket** stands in a **suburban driveway at golden hour**, arms folded, staring directly into the lens. She says, with rising frustration: "No. I'm not saying that. Write something else." She holds the stare, unmoving. Warm low sun, photoreal, locked-off camera. Suburban evening ambience, distant dog, no music. No on-screen captions, no zoom.

TweakRefusal is the sharpest variant of the trend. Keep the camera locked-off — any movement softens the confrontation.

Veo / Prompt Theory / self-aware AI characters

The pause before the loop

Open in builder
Prompt
Static wide shot of a **man sitting alone on a train**, window light moving across his face. He stares ahead, then turns to camera and asks calmly: "How many times have we done this?" He turns back to the window as if nothing happened. Photoreal, natural daylight, locked-off camera. Train rumble and rail clatter, no music, no captions.

TweakReturning to normal after the line is what makes it unnerving. State the return explicitly — Veo will otherwise hold the look to camera.

Veo / Prompt Theory / self-aware AI characters

Group scene — one person notices

Open in builder
Prompt
Medium wide shot of **four friends around a restaurant table**, warm evening light, mid-conversation and laughing. **The woman on the left** stops laughing first, looks around at the others, then straight into the lens. The others keep talking as if nothing changed. Photoreal, practical restaurant lighting, shallow depth of field. Restaurant ambience and overlapping chatter, no music.

TweakThe others continuing normally is the point. Say so, or Veo will have the whole table react.

Veo / Prompt Theory / self-aware AI characters

Direct address — no scene, just the look

Open in builder
Prompt
Tight close-up on the face of a **man in his fifties, weathered, unshaven**, against a plain dark background. He looks directly into the lens for a long beat, then says quietly: "You could stop reading at any time." Minimal soft key light from one side, deep shadow, photoreal, locked-off camera. Near silence, faint room tone, no music.

TweakStrip the setting entirely and the format gets much stronger. Give Veo one light source and nothing else to invent.

Veo / Prompt Theory / self-aware AI characters

Retail worker — the loop is the job

Open in builder
Prompt
Medium shot behind a **convenience store counter** under flat fluorescent light. A **bored young cashier** scans an item, then stops and says to the camera: "The customer resets too. You know that, right?" She resumes scanning. Photoreal, unflattering overhead fluorescents, locked-off camera. Barcode beeps, fridge hum, no music, no captions.

TweakUgly fluorescent light sells the realism here. Do not ask for cinematic grading — it breaks the found-footage feel.

Veo / Prompt Theory / self-aware AI characters

Vertical selfie variant (for Shorts / TikTok)

Open in builder
Prompt
Vertical 9:16 selfie video, handheld. A **woman in her twenties** walks through a **sunlit park**, talking casually to her phone camera mid-vlog. She stops walking, and her tone drops: "Who asked for this video?" She looks off to the side, then back to lens. Natural daylight, authentic phone-camera look, photoreal. Park ambience, footsteps, no music, no on-screen text.

TweakThe vlog framing gives the break somewhere to fall from. Keep the first half genuinely mundane.

Veo / Prompt Theory / self-aware AI characters

Interview subject questions the interviewer

Open in builder
Prompt
Seated interview framing, subject slightly off-centre, softly lit **home-office background** thrown out of focus. A **man in a knit sweater** finishes an answer, then leans forward and asks: "What happens to me when this clip ends?" He waits for an answer that does not come. Photoreal, soft key light, shallow depth of field, locked-off camera. Quiet room tone, no music, no captions.

TweakWaiting for a reply that never arrives is a strong closing beat. Write the wait into the prompt or Veo will cut on the line.

Veo / Prompt Theory / self-aware AI characters

Period setting — awareness out of time

Open in builder
Prompt
Medium shot of a **woman in 1940s dress** standing in a **train station concourse**, period crowd moving past in soft focus. She looks into the lens and says, puzzled: "This isn't when I am." Warm tungsten light, slight film grain, photoreal period detail, locked-off camera. Station announcements and crowd murmur, no music.

TweakPeriod settings make the anachronism of the awareness do extra work. Keep the crowd in soft focus so Veo does not have to resolve faces.

Veo / Prompt Theory / self-aware AI characters

The creator appears in frame

Open in builder
Prompt
Static medium shot of an **empty kitchen at night**, one light on over the counter. A **man walks into frame**, stops, and looks directly at the lens with recognition rather than surprise. He says: "Oh. It's you." Then he waits. Photoreal, single warm practical light, deep shadow, locked-off camera. House ambience, refrigerator hum, no music, no captions.

TweakRecognition instead of surprise is the whole beat — say "with recognition rather than surprise" explicitly, or Veo defaults to a startle.

Adapt any of these to your own shot

Every prompt above is a starting structure. To change the product, setting, camera move, or audio without breaking what makes it work, open it in the free Veo prompt builder — it rebuilds the prompt around your brief and gives you the Text and JSON versions side by side.

Open the Veo prompt builder

FAQ

What is Prompt Theory?

Prompt Theory is a viral AI-video format popularised on Veo 3 in which AI-generated characters behave normally and then break to camera, realising or refusing to accept that they are AI-generated. The videos are typically set in ordinary places — a street, a comedy club, a kitchen — and the whole effect rests on the contrast between the mundane setting and the break.

How do I make a Prompt Theory video in Veo?

Write a completely ordinary scene first, then add one short line delivered straight to the lens. The structure is: normal behaviour, a stop, direct eye contact with the camera, one line under about ten words, and a defined reaction afterwards. Keeping the setting mundane and the camera locked-off is what makes the break land.

Why does my character look at the camera the whole time?

Because you did not tell Veo what happens before and after the line. Describe the ordinary action first, then the turn to camera, then explicitly what the character does next — returns to the task, holds the stare, or waits. Without that, Veo holds the look to camera for the entire clip.

Do these prompts work on Veo 3.1 Lite?

Yes, though expect more re-rolls. The format depends on lip-sync and a controlled emotional beat, which are the two things the Lite tier approximates most. Draft on Lite or Fast to check the structure, then re-run the identical prompt on full Veo 3.1 for the take you keep.

Are these prompts verified?

Not yet. Every prompt on this page is labelled untested — they are written to the structure that the format relies on, but they have not been run against a real Veo output and no public evidence video is attached to them. Treat them as starting structures rather than proven results.