Updated September 23, 2026

How to Make an AI Found-Footage Home Video: Every Prompt Used

A fake 1997 Hi8 home video built with Midjourney, GPT Image and Seedance 2.5: the character still, reference sheet, four location plates and four clips, with every prompt in run order.

Free AI Video Hub TeamBy Free AI Video Hub Team9 min read

This is a found home video that never existed: one character, four locations, four silent clips, built so the whole reel reads as one tape someone dug out of a drawer. Every prompt in the pipeline is below, in the order it was run, with the settings and the failures that shaped them.

The method is the same at every step: lock one thing, then hand it to the next tool as a reference. A single face becomes a reference sheet, four empty locations become plates, and each clip opens on its plate's exact frame so the video model animates away from a fixed start instead of reinterpreting a description.

Part of our AI Video Generation Statistics 2026 series.

4 clips

One character, four locations and a single Hi8 tape look, all from 1 face, 1 reference sheet and 4 location plates.

Tools: Midjourney, GPT Image and Seedance 2.5.

Our take: The whole trick is the line "the opening frame matches this plate". It gives the model a fixed frame to animate away from, which is why four separate generations look like one tape.

Free AI Video Hub

01 / 5

The character

Everything downstream inherits this face. Generate a dozen, pick one, and never regenerate her again. The autumn street still is the only image where her face is invented rather than matched.

The Midjourney street still
Autumn street still. This is the only image where her face is invented rather than matched.
Street stillMidjourney1,076 ch
Cinematic 1990s 35mm film still, stylish woman, early twenties, walking Amsterdam canal streets in autumn. Tall, long voluminous golden blonde blowout, blue eyes, natural skin texture with visible pores. Fitted leopard print black velvet long sleeve top, ruched scooped neckline, high waisted dark indigo jeans, wide black leather belt with ornate gold western buckle and studs. Layered gold necklaces, thin chain choker, small gold hoops, black quilted chain strap shoulder bag, takeaway coffee in hand. Candid mid stride, glancing past lens. Background: narrow gabled canal houses with tall shuttered windows, arched stone bridge over the canal, rows of bicycles leaning on iron railings, moored houseboats and a glass tour boat, fallen orange elm leaves on wet cobblestones, blurred pedestrians. Warm late afternoon raking sunlight, long shadows, velvet catching directional sheen. 50mm lens, eye level, medium shot, shallow depth of field, fine grain, lifted blacks, warm nostalgic color grade, no retouching. --chaos 20 --ar 3:4 --raw --stylize 150 --hd --profile oi46hxe

Opens with this prompt already filled in. No signup.

02 / 5

The reference sheet

Five panels and seven macro swatches. This is what every clip points at, so the face holds across a whole reel instead of drifting shot to shot.

The character reference sheet
Front, three quarter, profile, back, close-up, plus material swatches along the bottom.
Reference sheetGPT Image1,579 ch
Create a character reference sheet of the woman from the street photo (long voluminous golden blonde blowout, blue eyes, layered gold necklaces), preserving her exact facial identity and likeness in every panel. Use the layout, panel structure and typography of the reference sheet with the grey studio background: a "CHARACTER REFERENCE SHEET" header, five full-height panels, and a bottom row of seven macro swatches. The five panels read left to right as front full body, three quarter turn, side profile, back view, and a head and shoulders close-up. Take the outfit entirely from the street photo: fitted leopard print black velvet long sleeve top with a ruched scoop neckline, high waisted dark indigo jeans, wide black leather belt with an ornate gold western buckle and gold studs, layered gold chain necklaces with a thin choker, small gold hoop earrings, black quilted leather bag on a gold chain strap, and black platform loafers. The seven macro swatches are leopard print velvet nap, indigo denim weave, black leather belt grain, the gold buckle and studs, quilted bag leather, gold chain links, and loafer leather. Shoot every panel on a plain warm grey seamless in an even soft studio key with gentle fill and a faint floor shadow, neutral to slightly warm color temperature, eye level 85mm lens for the full body views, 100mm for the close-up and macro swatches. Keep skin texture, pores, velvet nap, denim weave and hair flyaways fully visible with light 35mm film grain, no smoothing. Same person, same outfit and same lighting across all panels, ultra detailed

Opens with this prompt already filled in. No signup.

03 / 5

The location plates

Four empty sets, each framed from the exact camera position its clip will use. They are shot clean on purpose: baking tape noise into a still destroys the detail the video model needs to read geometry.

Vondelpark
Tulip field
Enclosed canal boat
Canal bridge

A · Vondelpark

Location plateMidjourney953 ch
Location reference plate, Vondelpark Amsterdam, early spring, mid afternoon, 1997, completely empty, no people in frame. Low camera sitting on the grass, about 40cm off the ground, looking slightly up: an open patch of green grass in the foreground, blossom branches in pale pink and white hanging into the top of the frame, bare and budding trees behind. Two old black Dutch city bikes lying dumped on their sides on a gravel path a few metres back, a rolled dark jacket and an open brown paper bag on the grass. Further back, a wide lawn, a pond edge and distant park benches, all out of focus. Hazy spring sun, soft blue haze, bright sky blowing out through the branches, dappled light on the grass. Seated low angle, 28mm lens, medium wide, interior of the scene fully sharp, shallow depth of field on the far background. Natural daylight only, neutral to slightly warm, no modern objects, no signage, no text. --ar 4:3 --raw --stylize 100 --chaos 0

Opens with this prompt already filled in. No signup.

B · Tulip field

Location plateMidjourney851 ch
Location reference plate, commercial tulip field in the Dutch countryside, April, 1997, completely empty, no people in frame. Standing camera at chest height looking down the rows: long dead straight rows of red and yellow tulips in solid blocks of colour running flat to the horizon, dark soil visible between the rows, a shallow drainage ditch along one edge. A line of bare poplars in the far distance, a low red brick farm shed with a corrugated roof, a flat pale grey blue sky with weak sun behind thin cloud. Wind laying the flowers over in visible waves. Standing eye level, 35mm lens, deep perspective down the rows, foreground and mid ground sharp, horizon soft. Flat diffused daylight, neutral colour, no windmill, no tourists, no buildings other than the shed, no modern vehicles, no signage, no text. --ar 4:3 --raw --stylize 100 --chaos 0

Opens with this prompt already filled in. No signup.

C · Enclosed canal boat

Location plateMidjourney1,263 ch
Location reference plate, interior of an enclosed glass-roofed Amsterdam canal boat moving slowly along a canal in spring, completely empty, no people in frame. View from the centre aisle at seated eye level looking down the length of the salon: a row of small polished oval wooden tables along each side, one narrow aisle between them, low tufted blue velvet banquette benches, varnished wood panelling. Panoramic windows running the full length of both sides from tabletop height up into a curved glass roof, varnished wood ribs between the panes, brass fittings, low varnished ceiling, worn wooden floor. Through the glass on both sides: narrow gabled canal houses with tall shuttered windows, moored houseboats, bicycles leaning on iron railings, trees just coming into leaf, a low arched stone bridge ahead, all out of focus. Soft daylight flooding in through the glass, faint reflections and a few rain beads on the panes, moving water bounce across the ceiling, no hard sun patches. Seated eye level, 28mm lens, symmetrical one point perspective down the aisle, interior fully sharp, shallow depth of field on the view outside only. Neutral to slightly warm colour, fine grain, no modern objects, no signage, no text. --ar 4:3 --raw --stylize 100 --chaos 0

Opens with this prompt already filled in. No signup.

D · Canal bridge

Location plateMidjourney942 ch
Location reference plate, narrow arched stone bridge over an Amsterdam canal, spring afternoon, 1997, completely empty, no people and no bicycles moving. Camera standing on the cobbled quay at the foot of the bridge, chest height, looking up the arch: worn cobblestones in the foreground, the bridge rising to a low crest with iron railings on both sides, canal water and moored wooden boats below on either side. Rows of old bikes chained along the railings. Gabled canal houses with tall shuttered windows lining both banks, trees just coming into leaf, a couple of parked 1990s European cars far back on the quay. Bright hazy spring daylight, bright blown out sky over the crest of the bridge, long soft shadows on the cobbles. Standing eye level, 35mm lens, medium wide, cobbles and bridge fully sharp, background soft. Neutral to slightly warm, no modern cars, no modern signage, no logos, no text. --ar 4:3 --raw --stylize 100 --chaos 0

Opens with this prompt already filled in. No signup.

04 / 5

The clips

Four silent takes on Hi8. Each one feeds its own plate as @img1 and the reference sheet as @img2, and each opens on the plate's exact frame so the model animates away from a fixed start instead of reinterpreting a description.

Clip 1 · Vondelpark · 9s

VondelparkSeedance 2.53,245 ch
REFERENCE MAP
@img1 = location: Vondelpark plate. Recreate this exact place, layout, set dressing, camera height and light. Do not redesign it. The opening frame matches this plate. Ignore that it is empty, she is lying on the grass in the foreground.
@img2 = character: face, long voluminous golden blonde hair, blue eyes, small gold hoops, layered gold necklaces with thin choker. Match face and hair exactly. Ignore the outfit on the sheet.

FORMAT
Amsterdam, spring, April 1997. Home video shot on a Hi8 camcorder by a friend sitting on the grass nearby. 4:3 frame. Small orange dot matrix date stamp "14 04 1997" bottom right, never moving. Soft video bloom on highlights, tape noise in the shadows, slight interlace combing on fast movement, auto-exposure hunting when she moves against the sky, mild red colour bleed, soft lens, faint grain from a later digital capture.
Handheld at the plate's camera height. Constant unsteady drift, never stabilized, never locked off. Any zoom is manual, clumsy, overshoots and corrects.
Outfit: fitted cream knit cardigan over a plain white top, light wash high waist jeans, small gold hoops, layered gold necklaces with a thin choker.
NO SPEECH. Nobody talks, no voices, no words, no lip movement forming words. Environment audio only.
Period accurate: no phones, no modern screens, no modern cars, no logos, no brand names, no text anywhere except the date stamp.

TIMELINE (9s, single take, no cuts)
0.0-2.5 She lies back on the grass with eyes closed, face to the sun, one knee up, hair spread out. Blossom moves overhead. Camera drifts lazily, framing slightly crooked, horizon tilted 3 degrees.
2.5-4.5 She lifts a hand to shield her eyes, squints, and finds the camera pointed at her. Auto-exposure hunts as her hand crosses the bright sky, the frame pumps brighter then darker.
4.5-7.0 She props up on one elbow, breaks into a real laugh with her mouth closed, and flicks her free hand at the lens to wave it away. No words, just the laugh and the gesture. Camera jolts back a little, guilty.
7.0-9.0 She flops back down and throws an arm over her face, still smiling underneath it. Camera does a clumsy zoom in, overshoots, pulls back and corrects. Ends mid-movement, no settle, no fade.

AUDIO
Environment only. Birdsong close and layered, wind thumping and popping the built-in mic, distant children and a dog, a bicycle bell somewhere off, the faint mechanical whir of the camcorder motor and tape transport. Her laugh is audible but wordless. No music, no narration, no other voices close to the mic.

CONSTRAINTS
No speech, no dialogue, no mouthed words, no voiceover, no music, no subtitles or on-screen text except the date stamp.
One laugh only, natural and brief. No shrieking, no shouting.
Nobody else enters frame. No hand reaching in, no second person, no background figures reacting or looking at camera.
No cuts, no stabilization, no gimbal smoothness, no slow motion, no speed ramp, no drone or crane move.
Camera stays at the plate's height on the grass. No overhead shot, no rising move, no relocating.
Do not change the location, the bikes, the blossom or the light from @img1.
Identity, hair and jewelry consistent with @img2. Outfit identical across all clips in this set.

Opens with this prompt already filled in. No signup.

Clip 2 · Tulips · 9s

TulipsSeedance 2.53,361 ch
REFERENCE MAP
@img1 = location: tulip field plate. Recreate this exact place, row layout, horizon, poplars, shed, sky and camera height. Do not redesign it. The opening frame matches this plate. Ignore that it is empty, she is crouched at the end of a row.
@img2 = character: face, long voluminous golden blonde hair, blue eyes, small gold hoops, layered gold necklaces with thin choker. Match face and hair exactly. Ignore the outfit on the sheet.

FORMAT
Netherlands, spring, April 1997. Home video shot on a Hi8 camcorder by a friend standing a few steps away. 4:3 frame. Small orange dot matrix date stamp "14 04 1997" bottom right, never moving. Soft video bloom on highlights, tape noise in the shadows, slight interlace combing on fast movement, auto-exposure hunting against the bright sky, mild red colour bleed that smears the tulip reds, soft lens, faint grain from a later digital capture.
Handheld at the plate's camera height. Constant unsteady drift, never stabilized, never locked off. Focus hunts badly on close objects before settling. Any zoom is manual and clumsy.
Outfit: fitted cream knit cardigan over a plain white top, light wash high waist jeans, small gold hoops, layered gold necklaces with a thin choker.
NO SPEECH. Nobody talks, no voices, no words, no lip movement forming words. Environment audio only.
Period accurate: no phones, no modern screens, no modern cars, no logos, no brand names, no text anywhere except the date stamp.

TIMELINE (9s, single take, no cuts)
0.0-2.0 She is crouched at the end of a row, cardigan sleeves pulled over her hands, looking down the stripes of colour. Wind lifts her hair across her face. Camera drifts, horizon tilted, the red rows blooming and bleeding slightly in the tape.
2.0-4.5 She picks a single tulip, stands, and brings it right up to the lens, turning the stem between her fingers. Focus hunts hard, goes soft, hunts again, finally locks on the petals. The frame fills with red for a moment.
4.5-7.0 She pulls it back from the lens, tucks it behind her ear above the gold hoop, and looks up with a closed mouth grin, pleased with herself. Exposure pumps as her face moves against the bright sky.
7.0-9.0 She turns and walks a few steps away down the row, then glances back once over her shoulder. Camera follows late and overshoots, catching empty rows before finding her again. Ends mid-movement, no settle, no fade.

AUDIO
Environment only. Strong steady wind across an open field, the mic clipping and rumbling on the gusts, dry rustle of leaves and stems, a distant tractor, a skylark high up, the faint whir of the camcorder motor and tape transport. No music, no narration, no voices.

CONSTRAINTS
No speech, no dialogue, no mouthed words, no voiceover, no music, no subtitles or on-screen text except the date stamp.
She picks one flower only. No bouquet, no handfuls, no trampling the rows.
Nobody else enters frame. No hand reaching in, no second person, no tourists. No windmill appears.
No cuts, no stabilization, no gimbal smoothness, no slow motion, no speed ramp, no drone or aerial move, no crane.
Camera stays at the plate's height on the ground. No overhead shot of the field, no relocating.
Do not change the row layout, the horizon, the shed or the sky from @img1.
Identity, hair and jewelry consistent with @img2. Outfit identical across all clips in this set.

Opens with this prompt already filled in. No signup.

Clip 3 · Canal boat · 10s

Canal boatSeedance 2.54,291 ch
REFERENCE MAP
@img1 = location: enclosed wooden canal boat plate. Recreate this exact boat, polished oval wooden table, tufted blue benches, wood panelling, windows and camera position. Do not redesign it. The opening frame matches this plate. Ignore that it is empty, she is sitting on the bench behind the table facing the camera. One green glass soda bottle and a small glass sit on the table, nothing else added.
@img2 = character: face, long voluminous golden blonde hair, blue eyes, small gold hoops, layered gold necklaces with thin choker. Match face and hair exactly. Ignore the outfit on the sheet.

FORMAT
Amsterdam, spring, April 1997. Home video shot on a Hi8 camcorder by a friend sitting across the table. 4:3 frame. Small orange dot matrix date stamp "15 04 1997" bottom right, never moving. Soft video bloom on highlights, tape noise in the shadows, heavy interlace combing on fast movement, violent auto-exposure hunting when the light changes, mild red colour bleed, soft lens, faint grain from a later digital capture.
Handheld at the plate's camera height. Constant unsteady drift plus a separate slower boat sway underneath it, two different rhythms. Never stabilized, never locked off. Any zoom is manual and clumsy.
Outfit: fitted cream knit cardigan over a plain white top, light wash high waist jeans, small gold hoops, layered gold necklaces with a thin choker.
NO SPEECH. Nobody talks, no voices, no words, no lip movement forming words. Environment audio only.
Period accurate: no phones, no modern screens, no logos, no brand names, no text anywhere except the date stamp.

TIMELINE (10s, single take, no cuts)
0.0-2.5 She sits behind the polished table, forearms resting on it, watching out of the side window. One window is pushed partly open and a thin draught keeps lifting a strand of her hair. Canal houses slide past the glass behind her with real parallax, their reflections drifting across the pane and over her face. Highlights bloom and smear on the polished tabletop and the brass fittings.
2.5-5.0 She hooks the loose strand back behind her ear, grins with her mouth closed, and points out of the window past the camera at something ahead. Camera swings that way late, catches nothing but blurred houses, swings back and overshoots her.
5.0-6.5 The boat passes under a low bridge. The cabin drops into shadow, auto-exposure hunts hard, the image goes murky and grainy, then blows out bright as it comes out the far side. Engine noise booms and echoes off the stone through the hull.
6.5-10.0 She has ducked down behind the table laughing silently, hair falling forward. She sits back up, picks up the green glass bottle and lifts it toward the lens in a small silent cheers, then takes a sip. Camera dips and recovers. Ends mid-movement, no settle, no fade.

AUDIO
Environment only, everything slightly muffled by the glass and the hull. Low inboard engine thrumming through the wooden floor, water slapping the hull, faint creak of timber, the boat's own rattle, gulls and a bicycle bell coming through thin and distant from outside, the engine booming and echoing under the bridge then opening out again, faint city hum, the whir of the camcorder motor and tape transport. Her laugh is audible but wordless. No music, no narration, no other voices.

CONSTRAINTS
No speech, no dialogue, no mouthed words, no voiceover, no music, no subtitles or on-screen text except the date stamp.
Enclosed wooden canal boat only. No open boat, no gondola, no large modern tour vessel.
She is alone. No second passenger, no crew, no hand entering frame, no one visible on the quay reacting.
No reflection of the camera or the operator ever visible in the windows or the polished table.
No cuts, no stabilization, no gimbal smoothness, no slow motion, no speed ramp, no drone or aerial move.
Camera stays at the plate's position across the table. It never relocates, never moves to another seat.
The boat never stops, never turns sharply, never rocks violently. Nothing falls off the table.
The bridge passage is one smooth light change. No flicker, no strobing, no fade to black, no cut.
Do not change the boat, its table, benches, panelling or fittings from @img1.
Identity, hair and jewelry consistent with @img2. Outfit identical across all clips in this set.

Opens with this prompt already filled in. No signup.

Clip 4 · Bike on the bridge · 9s

Bike on the bridgeSeedance 2.53,513 ch
REFERENCE MAP
@img1 = location: canal bridge plate. Recreate this exact bridge, arch, railings, cobbles, houses and camera position on the quay. Do not redesign it. The opening frame matches this plate. Ignore that it is empty, she appears at the top of the arch.
@img2 = character: face, long voluminous golden blonde hair, blue eyes, small gold hoops, layered gold necklaces with thin choker. Match face and hair exactly. Ignore the outfit on the sheet.

FORMAT
Amsterdam, spring, April 1997. Home video shot on a Hi8 camcorder by a friend standing on the quay. 4:3 frame. Small orange dot matrix date stamp "15 04 1997" bottom right, never moving. Soft video bloom on highlights, tape noise in the shadows, heavy interlace combing as she rides past, auto-exposure hunting against the bright sky over the bridge, mild red colour bleed, soft lens, faint grain from a later digital capture.
Handheld at the plate's camera height, standing still then turning on the spot to follow her. Constant unsteady drift, never stabilized, never locked off. The pan is clumsy and human: it lags, overshoots, loses her, finds her again.
Outfit: fitted cream knit cardigan over a plain white top, light wash high waist jeans, small gold hoops, layered gold necklaces with a thin choker. She rides an old black Dutch city bike with an upright frame and a chain guard.
NO SPEECH. Nobody talks, no voices, no words, no lip movement forming words. Environment audio only.
Period accurate: no phones, no modern screens, no modern cars, no helmets, no logos, no brand names, no text anywhere except the date stamp.

TIMELINE (9s, single take, no cuts)
0.0-2.5 She appears at the top of the arch, pushes off, and freewheels down toward the lens, cardigan lifting, hair going back. Exposure hunts as she crosses the bright sky at the crest.
2.5-5.5 She rides straight at the camera, one hand on the bars, and rings the bell twice, then lifts her other hand in a big loose wave. Closed mouth grin, no words. Heavy combing on the wheels and spokes.
5.5-7.5 She passes very close on frame right. The operator turns clumsily to follow, loses her entirely for a moment on empty railings and water, then swings back and catches her from behind.
7.5-9.0 She rides away along the quay, stands up on the pedals, and looks back once over her shoulder. Camera holds, still drifting, losing her slightly to the edge of frame. Ends mid-movement, no settle, no fade.

AUDIO
Environment only. Bicycle bell ringing twice and close, tyres drumming on cobbles, chain and freewheel ticking, wind across the mic as she passes, canal water, gulls, a distant tram bell and low city hum, footsteps and clothing rustle from the operator turning on the spot, the whir of the camcorder motor and tape transport. No music, no narration, no voices.

CONSTRAINTS
No speech, no dialogue, no mouthed words, no voiceover, no music, no subtitles or on-screen text except the date stamp.
She stays on the bike the whole clip. No dismount, no wobble, no fall, no near miss, no collision.
Nobody else enters frame. No second cyclist, no pedestrians crossing close, no hand reaching in.
No cuts, no stabilization, no gimbal smoothness, no slow motion, no speed ramp, no drone or aerial move, no tracking vehicle.
Camera stays standing on the quay and turns on the spot only. It never travels, never relocates.
Do not change the bridge, railings, cobbles or houses from @img1.
Identity, hair and jewelry consistent with @img2. Outfit identical across all clips in this set.

Opens with this prompt already filled in. No signup.

05 / 5

What breaks, and the fix

Every one of these came out of an actual failed render. Reach for the fix before you burn four more generations on the same beat.

WhereFailureFix
Any clipA voice appearsKeep her mouth closed in every smiling beat. Open mouths are where the model decides to generate speech.
Clip 4Spokes and hands meltBikes are the weakest geometry these models handle. Cut the timeline to the first two beats and end on the wave.
Clip 3The bridge reads as a cutSay the exposure "rolls down as the arch fills the windows" rather than that the frame goes dark.
Clip 3Hands mangle the bottleDrop the bottle and the cheers. End on her coming up from behind the table, both palms flat on the wood.
Clip 2Focus never settles on the flowerFocus hunting is scripted on purpose here. If it never resolves, cut "hunts again" and leave one hunt.
Character stillFlagged as NSFWRemove figure and bust descriptors. They read as a cluster, not as individual words.
All platesFour different placesChaos 0. Variations should be four angles on one location, not four locations.

Ready to make your own tape?

Take any prompt above, swap in your own character and locations, and generate it with the model of your choice.

Frequently asked questions

How do you keep the same character across AI video clips?

Generate one face, build a five-panel reference sheet from it, and pass that sheet into every clip as a reference image. In this pipeline the sheet goes in as @img2 alongside a location plate as @img1, so the face and hair stay fixed while the location and action change.

What is a location plate?

An empty still of a set, framed from the exact camera position a clip will use. Each clip opens on its plate's frame, so the video model animates away from a fixed start instead of reinterpreting a written description of the place.

How do you get the Hi8 home video look?

Keep the stills clean and put the tape look in the video prompt only: a dot matrix date stamp, highlight bloom, interlace combing, auto-exposure hunting and a handheld camera that drifts and overshoots. Baking noise into the plates destroys the detail the video model needs to read geometry.

Why do AI video clips sometimes add speech you did not ask for?

Open mouths are where the model decides to generate a voice. Keep her mouth closed in every smiling beat, and state no speech, no lip movement forming words and environment audio only in both the format and constraints sections.