AI Video Prompts for Horror: 10 Found-Footage Scares
Horror AI video prompts for found footage, liminal spaces, trail cams and CCTV scares. Copy-ready scenes built on dread, sound and a late reveal.
The horror AI video prompts that work in short form are almost never about monsters. They are about a camera recording something ordinary for a few seconds too long: a hallway that goes on too far, a trail cam that takes one extra photo, a figure in the background of a home video that nobody mentions. Below are 10 copy-ready scenes for found footage, liminal spaces and surveillance scares, plus the structure behind them so you can write your own.
Why Horror Needs Its Own Prompt Approach
Horror breaks most of the usual prompting advice. More detail about the subject usually hurts, and a beautiful, well-lit frame kills the mood. Four things make horror clips different:
- The camera is the narrator. A VHS camcorder, a night-vision baby monitor, a trail cam and a hotel CCTV each carry a different kind of fear. Pick the device first and describe its flaws.
- Timing lives inside the prompt. The scare has to arrive late. If you describe the figure in the first sentence, the model shows it in the first frame. Write the normal part first and use words like “then,” “after a few seconds” and “in the final moment.”
- Sound does half the work. Hum, silence, a single knock. Models with native audio (Veo 3.1, Kling 3.0, Seedance 2.0, Hailuo H3) can generate this, so write it in.
- Filters reward implication. Graphic violence and gore get blocked on mainstream models. The scariest short horror doesn’t need them.
The Horror Prompt Formula
[Recording device + its flaws] + [place with ONE wrong detail]
+ [ordinary action first] + [the anomaly, delayed]
+ [how the camera reacts] + [sound: ambience, then the break]
+ [ending: cut, glitch or freeze]
The “one wrong detail” is the key. An empty mall is a mood. An empty mall where every escalator is still running is a story.
Ready-to-Use Horror AI Video Prompts
1. Backrooms VHS Walk
Recovered 1990s VHS camcorder footage, first-person handheld walking
POV, burned-in date stamp in the corner. Endless maze of rooms with
mono-yellow wallpaper, stained damp carpet and buzzing fluorescent
ceiling panels. The camera walks slowly, turns a corner, then another
identical corner. In the final seconds, far down a corridor, a tall
thin shape steps out of view. The camera stops. Tape grain, chroma
bleed, slight tracking wobble. Constant fluorescent hum, soft
footsteps on wet carpet, then the hum cuts out completely.
Why it works: This is the yellow-room Backrooms look that started as a 2019 forum image, and the repeating corners build unease before anything appears. The sudden silence is the scare.
2. Liminal Mall at 3 AM
Handheld phone video, night mode, slight noise in the shadows. An
empty shopping mall after closing, lights on, every store shutter
down, soft elevator music playing from ceiling speakers. The camera
walks past a fountain that is still running. Every escalator in the
atrium is moving, including one that is running downward with a
single shopping bag riding it. No people anywhere. Echoing music,
trickling water, the phone holder's breathing getting faster.
Why it works: Liminal horror is about places built for crowds with nobody in them. The lone shopping bag is the wrong detail that implies someone was just there.
3. Trail Cam Sequence
Infrared trail camera footage in a pine forest at night, black and
white, fixed wide angle, timestamp and temperature readout at the
bottom. A deer walks through the clearing and leaves. The frame
stays empty for a few seconds. Then something pale and upright
crouches at the edge of the trees, watching the camera, completely
still. The footage stutters like a motion-triggered capture. Crickets,
then total silence.
Why it works: The deer establishes that the camera is real and normal. Stillness is scarier than movement, and “motion-triggered stutter” copies how real trail cams record.
4. Hotel Hallway CCTV
CCTV security camera footage, high ceiling angle looking down a long
hotel corridor with patterned carpet and identical doors, greenish
fluorescent lighting, low frame rate, timestamp overlay. A guest
walks away down the hall and enters a room. A few seconds later, the
door at the far end opens by itself, slowly, and stays open. The
hallway lights flicker once. Static surveillance, compression
artifacts. Faint electrical buzz, a distant door click.
Why it works: Surveillance footage feels objective, so small events land hard. Nothing supernatural is shown, only a door behaving wrong.
5. Baby Monitor Night Vision
Night-vision baby monitor footage, grainy grey and white, fisheye
lens, fixed high corner angle over an empty nursery with a crib and
a rocking chair. A stuffed rabbit sits in the rocking chair. The
chair begins to rock gently on its own. The rabbit's head slowly
turns toward the camera. Monitor static crackles. Soft lullaby from
a music box that slows down and stops.
Why it works: The monitor frame is instantly recognizable and domestic. Keeping the crib empty avoids generating a real child and keeps the focus on the toy.
6. Dashcam Empty Road
Dashcam POV footage at night on a two-lane rural road, headlights
lighting the asphalt, speed readout and timestamp on screen, wide
lens distortion. Trees pass on both sides. A figure in a long coat
stands on the shoulder of the road, facing away. The car passes it.
Ten seconds later the same figure is standing on the shoulder again,
still facing away. Engine drone, tires on asphalt, a radio playing
static.
Why it works: Repetition is a classic dread device. The viewer notices before the driver reacts, which makes them rewatch, and rewatches help retention.
7. Elevator Floor That Shouldn’t Exist
Handheld phone video inside an old apartment elevator, fluorescent
light, scratched metal walls. The floor display counts down 5, 4, 3,
2, 1, then shows a blank dash and the elevator keeps descending. The
doors open onto a dark concrete corridor lit by a single red bulb.
The phone camera shakes slightly. Mechanical grinding, a ding, then
dripping water and a low distant rumble.
Why it works: The floor counter is a visual clock that makes the viewer count along. Numbers on a simple display usually render better than signs or long text, but check it before posting.
8. Mirror Delay
Selfie-style phone video in a small bathroom lit by a single warm
bulb. A young man in a grey hoodie brushes his teeth while looking
into the mirror. He stops and tilts his head. His reflection keeps
brushing for a moment, then slowly stops and tilts its head a
second later. Camera held steady at chest height. Running tap, the
brushing sound continues for a beat after he stops.
Why it works: The audio mismatch (brushing continues after he stops) sells the scare even if the visual delay is subtle.
9. Analog Horror Instructional Tape
1980s educational VHS tape, washed-out colors, 4:3 frame with black
borders, tracking lines rolling through the image. A smiling
presenter in a beige sweater stands in a bright suburban living room
next to a chalkboard that reads "LESSON 4". She explains something
cheerfully, then stops smiling and stares at the camera in silence
while the background room slowly darkens. Tape hiss, cheerful synth
music that warps and slows.
Why it works: Analog horror plays on friendly formats turning wrong. Short, simple board text is more likely to render than a full sentence.
10. Poolrooms
Handheld camcorder footage, soft bright light, washed-out colors,
slight overexposure. An endless indoor swimming pool complex of
white tiles, shallow still water and arched doorways leading to more
pools. The camera wades slowly through ankle-deep water. Somewhere
ahead, ripples arrive from a doorway where nobody is visible. The
camera stops. Echoing water drips, a distant splash, heavy breathing.
Why it works: Bright liminal horror is less common than dark hallways, so it stands out in a feed. The ripples are the only evidence of the presence.
Model-Specific Tips for Horror
Veo 3.1 generates native audio synced to what’s on screen, which matters more in horror than anywhere else. Clips run 4, 6 or 8 seconds, so keep one setup and one reveal. The Veo 3 prompt generator page covers audio phrasing.
Kling 3.0 makes clips of up to 15 seconds and can cut between up to 6 shots in a single generation. Use that for escalation: wide establishing shot, closer shot, then the reveal. See the Kling prompt generator.
Seedance 2.0 accepts video references. Record a real walk through your hallway on your phone and use it to copy authentic handheld movement. The Seedance prompt generator has examples.
Hailuo H3 (MiniMax’s July 2026 model, sold as Hailuo 3.0) adds native stereo audio and clips up to 15 seconds. The Hailuo prompt generator covers the details.
For grading the look (crushed blacks, cold greens, tape wash) see our color grading prompts, and for VHS and camcorder eras, the vintage film guide.
Common Mistakes With Horror Prompts
- Revealing the threat in sentence one. Describe the normal moment first, the anomaly last.
- Cinematic lighting. Soft key lights and shallow depth of field look like a movie trailer, not footage. Found footage wants flat, harsh or failing light.
- Asking for gore. It gets filtered, and the clip usually wasn’t scarier with it.
- A clean camera. Real devices have noise, compression, focus hunting and timestamps. Leave them out and the clip looks rendered.
- Using real places. In August 2026, police in Ireland and Scotland had to debunk viral AI doorbell-style clips of a menacing Cat in the Hat figure. Keep locations generic and label the clip as AI.
- No ending. Decide the last beat: a hard cut, a glitch, a freeze. Our negative prompts guide helps when a model keeps adding unwanted people or bright light.
How LzyPrompt Generates These
Found-footage is the house style of LzyPrompt. Type a plain idea like “trail cam catches something standing in the trees” and you get 10 variations, each with a scene title, a full prompt that includes the camera device, lighting, action beats, device imperfections and sound cues, plus a ready-to-post caption. Real outputs already use camera styles like CCTV security camera, doorbell camera, dashcam POV and handheld phone video, so name the device in your idea or set the optional Camera Style dropdown (it defaults to Auto-detect).
The prompts are plain text for any model, and you can export all 10 to CSV when you’re building a week of horror Shorts in one sitting. Try LzyPrompt free for 7 days.
Frequently Asked Questions
Why does my horror prompt get blocked by the AI video tool?
Mainstream models run safety filters for violence and graphic content, so blood, injuries and gore are the usual triggers. Horror that works in short form rarely needs them anyway. Rewrite the scare as implication: a figure that is standing too still, a door that was closed a second ago, a sound with no source. Those pass filters and tend to be scarier.
Which AI video model is best for horror clips?
Veo 3.1 is strong because horror depends on sound, and it generates ambience, footsteps and silence breaks natively. Kling 3.0 can cut between up to 6 shots in one generation, which suits a slow escalation. Seedance 2.0 lets you upload a real walkthrough video as a motion reference, useful for convincing found-footage camera movement.
How do I make AI footage look like real found footage?
Name the recording device and its flaws: VHS tape grain, chroma bleed, a burned-in date stamp, autofocus hunting, auto-exposure pumping when the light changes, a camcorder mic that clips on loud sounds. Then ask for an ordinary action before the anomaly. Real found footage is boring for the first few seconds, and that is what makes the scare believable.
Is it okay to post realistic AI horror clips as if they were real?
Label them. In August 2026, police in Ireland and Scotland had to publicly debunk AI doorbell-style clips of a sinister Cat in the Hat figure that spread on TikTok. YouTube also requires creators to disclose realistic altered or synthetic content. Keep the location fictional, never name a real street or school, and use the platform's AI label.
How long should an AI horror short be?
Most models generate 8 to 15 seconds per clip, which is enough for one setup and one reveal. For a longer story, chain three or four clips that share the same camera device and location description, and put the scariest beat in the last clip.
Related Prompt Guides
AI Video Prompts for CCTV Security Camera Videos
CCTV AI video prompts for the animals-caught-on-camera trend: high corner angles, timestamps, IR night vision, choppy frames and ambient audio.
View prompts →AI Video Prompts for Halloween 2026: 10 Spooky Clips
Halloween AI video prompts for October 2026: costumed pets, haunted doorbell cams, pumpkin loops and spooky small-business ads. Copy, paste, post.
View prompts →AI Video Prompts for Realistic Handheld Phone Videos
Realistic phone video AI prompts that don't look AI: hand wobble, autofocus hunting, exposure shifts, phone-mic audio and loose vertical framing.
View prompts →Generate Custom Prompts with LzyPrompt
Stop writing prompts from scratch. Describe your scene, pick a camera style or let it auto-detect, and get 10 structured variations with captions. Paste them into Veo, Kling, Seedance, or any AI video tool.
Start Free Trial