POV phone footagevideo prompts

Ten prompts for AI video that looks like it was filmed on a phone: first-person, selfie camera, one continuous take. Each one was run on Veo 3.1 Fast and Seedance 2.0 on October 27, 2026. Below every clip is the exact prompt and what happened when we ran it.

6 of 10 nailed it on the first runThe rest got one rewrite. Each prompt below says what changed.

What the working prompts have in common

A named camera
First-person POV or front-camera selfie, held by someone. Not “cinematic”.
One continuous take
Cuts break the phone illusion, so every prompt asks for one shot.
Three timed beats
0–3s setup, 3–6s the turn, 6–8s the reaction. Models follow timestamps well.
Phone flaws
Grain, exposure hunting, raindrops on the lens. Perfect images read as AI.
Sound and one line
Ambient noise plus one short spoken line, written out word for word.
An avoid line
Name what would ruin it: visible ghosts, music, slow motion.

AI-generated with Veo 3.1 Fast. Unedited, sound on.

V01

The Bedroom Door

Veo 3.1 Fast, text-to-video, 9:16, 8s, 1080p, audio on

Rewritten once, this one works

Prompt
Vertical 9:16, 8 seconds, first-person POV handheld smartphone video, one continuous take, no cuts.
Night, a narrow apartment hallway lit only by the phone's flashlight. At the end of the hallway is a bedroom door that is fully closed.
0-3s: The camera walks slowly toward the closed door; the beam sways with each step, the floorboards creak, dust floats in the light.
3-6s: The camera stops two meters from the door. The door slowly swings open about two inches by itself. Pure darkness in the gap.
6-8s: The flashlight beam trembles on the gap; a quiet, sharp intake of breath from behind the camera. Nothing comes out.
Look: real phone night footage, heavy grain, auto-exposure hunting, slight motion blur, no color grading.
Audio: footsteps, creaking wood, a faint low hum, breathing. No music.
No people or figures appear at any point. The camera never goes through the door.

What went wrong first

First run had a person stepping into frame at the end. Adding the last line (no people, camera never goes through the door) fixed it on the next run.

AI-generated with Seedance 2.0 (eco-mini). Unedited, sound on.

V02

Who's Holding My Hand

Seedance 2.0 (eco-mini), text-to-video, 9:16, 8s, 720p, audio on

Rewritten once, this one works

Prompt
Vertical 9:16, 8 seconds, first-person POV smartphone video held in the right hand, one continuous take.
A dim bedroom at night, the viewer sits on the edge of the bed. Across the room, a young adult woman in a yellow sweater stands in the lit doorway, waving at the camera with BOTH hands raised high, smiling.
0-3s: The camera is pointed at the woman in the doorway; both of her hands are clearly up in the air, waving.
3-6s: The camera slowly tilts down to the viewer's own left hand resting on the bed sheet. A different hand is holding it: a thin, pale grey bare arm with no sleeve, reaching up from the dark gap under the bed.
6-8s: The pale fingers tighten around the viewer's hand. The camera freezes on it.
Look: authentic phone footage, warm light only from the hallway, grain in the shadows, handheld micro-shake.
Audio: the woman says softly "Hey, I'm home", then silence, a sheet rustling.
Avoid: gore, monster faces, fast cuts. The holding hand must not wear any sleeve.

What went wrong first

Veo 3.1 dressed the hidden hand in the woman's yellow sleeve, so the twist read as her own hand. Seedance followed "BOTH hands raised" and "no sleeve" exactly. The audio came out very quiet: boost it in editing.

Structure inspired by @john87445528

AI-generated with Seedance 2.0 (eco-mini). Unedited, sound on.

V03

The Elevator Mirror

Seedance 2.0 (eco-mini), text-to-video, 9:16, 8s, 720p, audio on

Rewritten once, this one works

Prompt
Vertical 9:16, 8 seconds, front-camera selfie video, one continuous take. The phone itself is never visible; we see what the front camera sees.
A young adult man stands in a small old elevator, facing the camera. Directly behind him is a mirror wall, which normally shows the back of his head.
0-3s: He looks into the camera, tired. The mirror behind him shows the back of his head, as expected.
3-5s: The ceiling light flickers off and on once. Now the mirror behind him shows his FACE instead of the back of his head: his reflection is looking straight at the camera over his shoulder, smiling slightly. He has not noticed and keeps looking at the camera with a tired expression.
5-8s: The light flickers again. The mirror is back to normal, showing the back of his head. He frowns slightly, as if he felt something.
Look: phone front camera, slightly soft focus, flicker banding, greenish fluorescent tint.
Audio: elevator hum, the light buzzing, a click on each flicker.
Avoid: distorted faces, horror makeup, text overlays.

What went wrong first

The hardest one. Veo turned it into a mirror selfie, and the first Seedance try had the reflection turning with him. The light-flicker version works, but the face in the mirror is small, so crop in if you want a bigger scare.

AI-generated with Veo 3.1 Fast. Unedited, sound on.

V04

Crying-Laughing in the Car

Veo 3.1 Fast, text-to-video, 9:16, 8s, 1080p, audio on

Worked on the first run

Prompt
Vertical 9:16, 8 seconds, front-camera selfie video, one continuous take, phone propped on the steering wheel.
A young adult woman in a parked car at dusk, mascara slightly smudged, laughing and wiping tears at the same time.
0-3s: She covers her mouth with one hand, shaking her head, half laughing.
3-6s: She looks into the camera and says "I got the job. I actually got the job."
6-8s: She leans back against the headrest, exhales, and laughs again with wet eyes.
Look: real phone selfie, warm orange streetlight through the windshield, slight lens smudge, natural skin texture, no beauty filter.
Audio: her voice, a car passing outside, the turn signal ticking once.
Avoid: dramatic music, perfect makeup, cinematic depth of field.

Test notes

Worked first try. The emotion reads well; the car interior is soft, so add "steering wheel visible" if the setting needs to be obvious.

Structure inspired by @john87445528

AI-generated with Veo 3.1 Fast. Unedited, sound on.

V05

Rainy Walk to the Subway

Veo 3.1 Fast, text-to-video, 9:16, 8s, 1080p, audio on

Worked on the first run

Prompt
Vertical 9:16, 8 seconds, front-camera selfie video while walking, one continuous take.
A young adult man in a dark raincoat walks down a rainy city street at 7 a.m., umbrella in his other hand, neon shop signs reflected on the wet pavement behind him.
0-4s: He walks toward the camera, slightly out of breath, and says "Okay, day three of waking up early. Still alive."
4-8s: He turns the phone to show the subway entrance ahead, rain dripping from the umbrella edge into frame.
Look: real phone vlog, walking bob, raindrops on the lens, cool blue morning light, mild compression artifacts.
Audio: rain on the umbrella, traffic, his voice close to the mic.
Avoid: slow motion, stabilized cinematic glide, music.

Test notes

Worked first try. Raindrops on the lens and the walking bob sell the phone look.

AI-generated with Veo 3.1 Fast. Unedited, sound on.

V06

Grandma's Dumplings

Veo 3.1 Fast, text-to-video, 9:16, 8s, 1080p, audio on

Worked on the first run

Prompt
Vertical 9:16, 8 seconds, first-person POV smartphone video, one continuous take.
A small home kitchen at night, steam rising from a pot. An elderly woman with grey hair and an apron stands at the stove.
0-3s: The camera enters the kitchen; she turns around, surprised and happy.
3-6s: She scoops dumplings from the pot into a bowl and pushes the bowl toward the camera with both hands, saying "Eat while it's hot."
6-8s: The viewer's free hand reaches into frame to take the bowl; steam fogs the lens slightly.
Look: real phone footage, warm tungsten kitchen light, steam haze, slightly overexposed window, natural wrinkles and skin texture.
Audio: boiling water, her voice, a spoon tapping the bowl.
Avoid: polished food commercial look, slow motion.

Test notes

Worked first try. It opens on a wider shot than pure POV before the bowl comes toward the lens, which still feels natural.

AI-generated with Veo 3.1 Fast. Unedited, sound on.

V07

Coming Home to the Dog

Veo 3.1 Fast, text-to-video, 9:16, 8s, 1080p, audio on

Worked on the first run

Prompt
Vertical 9:16, 8 seconds, first-person POV smartphone video held low, one continuous take.
An apartment front door opens from the outside into a dim hallway.
0-3s: As the door opens, a golden retriever is already sitting right behind it, tail sweeping the floor.
3-6s: The dog jumps up, paws on the viewer's knees, licking toward the lens; the camera shakes.
6-8s: The viewer laughs off-camera, "Okay, okay, I missed you too."
Look: authentic phone footage, motion blur on the wagging tail, warm hallway light, fur detail.
Audio: keys jingling, excited whining, laughter.
Avoid: slow motion, studio lighting, extra legs or deformed paws.

Test notes

Worked first try. Paws and fur held up well, with no extra legs.

AI-generated with Veo 3.1 Fast. Unedited, sound on.

V08

Concert from the Crowd

Veo 3.1 Fast, text-to-video, 9:16, 8s, 1080p, audio on

Rewritten once, this one works

Prompt
Vertical 9:16, 8 seconds, raw handheld footage filmed with one arm raised above the heads of a crowd, one continuous take. The image fills the whole frame edge to edge.
A packed outdoor music festival at night. Far away on stage, a singer is a small silhouette in front of a giant LED wall glowing purple and orange.
0-4s: Hundreds of raised hands sway in front of the lens; confetti starts falling.
4-8s: The crowd jumps on the beat; the shot shakes up and down; laser beams sweep over the audience.
Look: raw amateur concert footage, blown-out highlights on the stage, heavy noise in the crowd, rolling shutter on the lasers. No screen recording, no status bar, no navigation buttons, no borders.
Audio: distorted clipped bass, crowd screaming, no clear vocals.
Avoid: recognizable artist faces, logos, band names, clean studio audio.

What went wrong first

Twice Veo rendered a phone screen recording (status bar, navigation buttons, black borders) when the prompt said "smartphone video". Describing it as raw handheld footage that fills the frame fixed it.

AI-generated with Veo 3.1 Fast. Unedited, sound on.

V09

Night Market Order

Veo 3.1 Fast, text-to-video, 9:16, 8s, 1080p, audio on

Worked on the first run

Prompt
Vertical 9:16, 8 seconds, first-person POV smartphone video, one continuous take.
A busy night-market food stall with a hot griddle, strings of bare bulbs overhead.
0-3s: The camera moves up to the stall; a middle-aged vendor flips sizzling scallion pancakes and nods.
3-6s: He folds one into a paper bag, smiles, and hands it toward the camera, saying "Careful, it's hot."
6-8s: The viewer's hand takes the bag; steam and oil sparkle in the bulb light.
Look: real phone footage, warm bulb light, smoke haze, handheld micro-shake, slight lens flare.
Audio: sizzling oil, crowd chatter, his voice.
Avoid: signage with readable brand names, glossy commercial look.

Test notes

Worked first try. Background signs come out as made-up characters, which is fine at this distance.

AI-generated with Veo 3.1 Fast. Unedited, sound on.

V10

Someone Calls Your Name in the Fog

Veo 3.1 Fast, text-to-video, 9:16, 8s, 1080p, audio on

Worked on the first run

Prompt
Vertical 9:16, 8 seconds, first-person POV smartphone video while hiking, one continuous take.
A mountain trail in thick fog, wet pine trees, visibility a few meters.
0-3s: The camera walks forward on the gravel path, breathing audibly.
3-6s: From somewhere behind, a distant voice calls "Hey... wait up!" The camera stops and turns around slowly.
6-8s: The trail behind is empty, only fog. The camera holds still.
Look: real phone footage, flat grey light, low contrast, fog diffusion, slight lens condensation.
Audio: gravel crunch, wind in the pines, the distant voice, heavy breathing.
Avoid: visible figures, creatures, dramatic music.

Test notes

Worked first try. Keep "no visible figures" or models like to add a silhouette in the fog.

Write your own

Blank template. Fill in the brackets.
Vertical 9:16, [duration] seconds, [first-person POV / front-camera selfie] smartphone video, one continuous take.
[Location + time + light source].
0-3s: [setup action].
3-6s: [the turn or reveal].
6-8s: [reaction, then hold].
Look: real phone footage, [grain / blur / exposure quirks], no color grading.
Audio: [ambient sound], [one short spoken line].
Avoid: [whatever would break the realism].

Questions

Which model should I use for phone-footage prompts?

In our tests, Veo 3.1 Fast handled everyday and warm scenes (vlogs, pets, food, family) on the first run and adds believable sound. For scenes that depend on a logical twist, like a hand that should not be there, Seedance 2.0 followed the instructions more closely.

Why does my AI video look like a screen recording?

When a prompt says "smartphone video", Veo 3.1 sometimes draws the phone itself: a status bar, navigation buttons and black borders. Describe the footage instead, for example "raw handheld footage that fills the whole frame", and add "no screen recording, no status bar".

How much did these cost to generate?

Through an API reseller in October 2026, an 8-second 1080p Veo 3.1 Fast clip cost about $0.16 and an 8-second 720p Seedance 2.0 eco-mini clip about $0.29. The whole page, including failed runs, cost under $3.

Can I use these prompts?

Yes. Copy and change them however you like. What you generate is governed by the terms of the model you use.