Video & audio

Create an abandoned-hospital AI vlog

Prepare the source assets, follow the supplied generation workflow, and inspect the result frame by frame before exporting.

Adapted from TUT-018, dated 2026-04-14.

What you'll make

A checked final video that follows the source workflow and is ready for export.

Create an abandoned-hospital AI vlog

Who this is for

Creators, marketers, founders, and small teams producing short AI-assisted videos from source media.

Step by step

Run the workflow

  1. 01

    The full workflow behind how I made this exact video

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Follow the shown setup, keep the source order unchanged, and run one controlled test.
    Source example for create an abandoned-hospital ai vlog.
    Use this visual to confirm the expected setup or result.
    Source example for create an abandoned-hospital ai vlog.
    Use this visual to confirm the expected setup or result.
    Source example for create an abandoned-hospital ai vlog.
    Use this visual to confirm the expected setup or result.
    Source example for create an abandoned-hospital ai vlog.
    Use this visual to confirm the expected setup or result.
    Source example for create an abandoned-hospital ai vlog.
    Use this visual to confirm the expected setup or result.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  2. 02

    Prepare the source material

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Capture as many reference images as you need. You do not need professional camera gear for your character.
    Source example for the full workflow behind how i made this exact video.
    Use this visual to confirm the expected setup or result.
    Source example for the full workflow behind how i made this exact video.
    Use this visual to confirm the expected setup or result.
    Source example for the full workflow behind how i made this exact video.
    Use this visual to confirm the expected setup or result.
    Source example for the full workflow behind how i made this exact video.
    Use this visual to confirm the expected setup or result.
    Source example for the full workflow behind how i made this exact video.
    Use this visual to confirm the expected setup or result.
    Source example for the full workflow behind how i made this exact video.
    Use this visual to confirm the expected setup or result.
    Source example for the full workflow behind how i made this exact video.
    Use this visual to confirm the expected setup or result.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  3. 03

    Open the image generator

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Create your scenes or environment with NanoBanana Pro
    2. Go to Google Flow and select "Create with Flow."
    3. Select + New project
    Source example for prepare the source material.
    Use this visual to confirm the expected setup or result.
    Source example for prepare the source material.
    Use this visual to confirm the expected setup or result.
    Source example for prepare the source material.
    Use this visual to confirm the expected setup or result.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  4. 04

    Enter your prompts to generate the images

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. You may receive free credits (varies by account).
    Prompt setup used for open the image generator.
    Check the prompt field, reference order, and visible settings.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  5. 05

    Create Character Cheatsheet

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Image Model: NanoBanana Pro
    2. create a character cheatsheet for this person, front, back ,full body, mid shot, wide shot. all on white background
    Expected result for enter your prompts to generate the images.
    Use this source result as the visual comparison point for the step.
    Expected result for enter your prompts to generate the images.
    Use this source result as the visual comparison point for the step.
    Expected result for enter your prompts to generate the images.
    Use this source result as the visual comparison point for the step.
    Expected result for enter your prompts to generate the images.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  6. 06

    Create The Scenes

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Create The Scenes prompt
    Raw iPhone night photo of a real abandoned hospital entrance, handheld, documentary realism, low light, dark, slightly underexposed, not cinematic, not CGI. A real condemned hospital entrance at night photographed by someone standing outside with an iPhone. The image should feel like an actual low-light smartphone photo taken quickly in a creepy place, with natural handheld framing, slight underexposure, visible low-light noise, soft shadow detail, mild blur from hand movement, and imperfect phone camera exposure. The scene should be dark enough that some areas fall into shadow naturally, with only limited detail visible in the deepest parts of the entrance. The hospital entrance is old, decayed, and neglected. Cracked concrete steps, broken glass doors, rusted metal frames, peeling paint, stained walls, damp surfaces, mold streaks, grime, exposed wiring, and weeds growing through gaps in the pavement. A damaged or half-broken hospital sign sits above the doorway with missing letters or faded text. The canopy over the entrance is cracked, sagging, or partially broken. Debris is scattered near the entrance, including broken tiles, old caution barriers, a rusted wheelchair frame, or abandoned hospital equipment parts. Lighting should feel ugly, weak, and realistic. Very dark overall image, underexposed shadows, faint cold streetlight, slight dirty fluorescent spill from somewhere deep inside, damp reflections on the ground, maybe a little fog or humid haze. The inside of the building should be mostly swallowed by darkness. Do not make the lighting dramatic or beautiful. It should feel like a real phone photo from an urban explorer in a dangerous abandoned place. Realism cues: shot on iPhone handheld smartphone photo low light dark image slightly underexposed natural image noise limited dynamic range imperfect exposure messy real-world framing documentary photo urban exploration snapshot amateur realism found footage feel Negative prompt: CGI, 3D render, Unreal Engine, Octane, cinematic, movie still, glossy surfaces, dramatic lighting, perfect symmetry, concept art, stylized horror, digital painting, illustration, anime, game scene, fake fog, overly clean decay, studio lighting, monsters, zombies, gore, blood, watermark, text

    Expected result for create character cheatsheet.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  7. 07

    Generate a reference image 1

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Generate a reference image prompt
    Raw iPhone night photo of a real abandoned hospital hallway, handheld, documentary realism, very dark, low light, slightly underexposed, no fluorescent lights, not cinematic, not CGI. A real abandoned hospital corridor photographed on an iPhone in near darkness. The image should look like an actual handheld smartphone photo taken by someone cautiously standing in the hallway at night, with natural messy framing, slight underexposure, visible low-light noise, limited dynamic range, minor motion softness, and imperfect focus in darker areas. It should feel like a real urban exploration photo, not a polished horror image. The hallway is long, narrow, empty, and deeply unsettling. Old hospital corridor with dirty cracked tile flooring, stained walls, peeling paint, water damage, mold spreading along the ceiling corners, exposed pipes, hanging broken ceiling panels, damaged wall fixtures, and old patient room doors lining both sides. Some doors are slightly open, some closed, some damaged, with pitch-black interiors beyond them. The far end of the hallway disappears into almost complete darkness. Scattered debris covers parts of the floor: broken ceiling pieces, paper scraps, dirt, abandoned medical clutter, rust stains, damp patches, and maybe an overturned wheelchair or broken trolley frame partly visible in shadow. The walls should feel wet, dirty, moldy, and long neglected. Some parts of the floor can be slightly reflective from moisture, but the overall image should remain dark and hard to read. Lighting should be extremely minimal and natural. No fluorescent light, no bright practical lights, no visible ceiling fixtures casting light. The corridor is mostly swallowed by darkness, with only faint ambient night light leaking in from a distant window, an open doorway far away, or weak moonlight filtering indirectly into parts of the hallway. Most of the scene should remain underexposed. Let details disappear naturally into shadow. The atmosphere should feel silent, humid, oppressive, and real. Realism cues: shot on iPhone handheld smartphone photo very dark slightly underexposed natural low-light noise limited phone dynamic range soft shadow detail messy real-world framing documentary photo urban exploration snapshot amateur realism found footage feel Negative prompt: fluorescent lights, ceiling lights, dramatic lighting, cinematic lighting, CGI, 3D render, Unreal Engine, movie still, glossy surfaces, stylized horror, perfect symmetry, concept art, illustration, painting, anime, game environment, fake fog, monsters, zombies, gore, blood, text, watermark

    Expected result for create the scenes.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  8. 08

    Generate a reference image 2

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Generate a reference image prompt
    A real abandoned hospital toilet photographed in near total darkness, documentary realism, very dark, heavily underexposed, no visible phone in frame, no light leaking, not cinematic, not CGI. A real condemned hospital restroom captured as a raw low- light photograph from the point of view of someone standing inside the room. The image should feel like an actual photo of the space itself, not a stylized horror scene and not a movie still. The framing is natural and slightly imperfect, with realistic low-light noise, limited dynamic range, crushed shadows, soft detail loss in dark areas, and strong underexposure. No phone, hand, flashlight, or person should be visible in the frame. The toilet room is filthy, decayed, and long abandoned. Old stained ceramic toilet bowls, cracked wall tiles, peeling paint, water damage, mold spreading along grout lines and ceiling corners, rusted metal fixtures, damp floor, dirt, grime, and scattered debris. The toilet stall doors are closed. They should look old, worn, slightly warped, stained, and damaged by moisture. Under one of the closed stall doors, a pair of bare human legs and feet is visible beneath the bottom gap. Only the lower legs and bare feet can be seen. The skin should look pale, lifeless, dirty, and unnaturally still, with a ghostlike horror feel similar to a supernatural horror scene. The feet are bare, slightly grimy, and planted motionless on the dirty floor, creating a disturbing presence without showing the full body. The legs should feel eerie, unnatural, and deeply unsettling, like something silently waiting inside the stall. The room should have no visible light source and no light leaking in from anywhere. No fluorescent lights, no moonlight, no doorway spill, no bright highlights, no dramatic horror glow. The image should feel almost completely dark, with only barely readable forms emerging from the blackness because of the camera sensor trying to capture the scene. Most of the room should be lost in shadow. Let details disappear naturally. The darkness itself should be the main atmosphere. The materials must look physically real: grimy ceramic, stained porcelain, dirty tile, rusted hinges, damp floor, peeling paint, black mold, and realistic neglect. The final result should look like a real severely underexposed photo of an abandoned hospital toilet taken in almost complete darkness. Realism cues: raw low-light photo documentary realism near total darkness heavily underexposed crushed shadows natural image noise limited dynamic range imperfect real-world framing urban exploration photo amateur realism found footage feel no phone visible no hands visible no visible light source Negative prompt: light leaking, window light, moonlight, doorway light, flashlight beam, fluorescent lights, dramatic lighting, cinematic lighting, CGI, 3D render, Unreal Engine, movie still, stylized horror, glossy surfaces, concept art, illustration, painting, anime, game environment, gore, blood, text, watermark

    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  9. 09

    Generate a reference image 3

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Generate a reference image prompt
    Raw low-light photo of a real abandoned hospital toilet entrance, documentary realism, very dark, heavily underexposed, no visible light source, not cinematic, not CGI. A real condemned hospital restroom entrance photographed in near total darkness. The image should feel like an actual raw low-light photo of the location itself, not a movie still and not stylized horror. The framing is natural and slightly imperfect, with realistic low-light noise, crushed shadows, limited dynamic range, and soft detail loss in the darkest areas. No phone, hand, flashlight, or person visible in frame. The entrance to the toilet is narrow, dirty, and deeply unsettling. A decayed restroom doorway opens into blackness inside an abandoned hospital. The door frame is chipped, stained, damp, and mold- ridden. Old peeling paint, cracked tiles around the entrance, rust stains, water damage, exposed edges, grime buildup, and warped surfaces. Above or beside the entrance, remnants of old restroom signage or broken fixtures may be faintly visible but unreadable. The corridor outside the toilet feels neglected and rotting. The doorway itself should look like a threshold into something unsafe and silent. Inside the entrance, only vague shapes of old toilet stall partitions or tiled walls should be barely visible through the darkness. The scene should remain mostly swallowed by blackness, with only minimal readable detail captured by the camera sensor. No visible light spilling out from inside. No fluorescent lights. No moonlight. No dramatic glow. The darkness should feel heavy and real. Materials must look physically real: damp cracked tile, rotting paint, stained concrete, mold, rust, grime, and realistic decay. The final image must look like a real severely underexposed photo taken in a derelict hospital restroom entrance. Negative prompt: CGI, 3D render, Unreal Engine, cinematic lighting, movie still, stylized horror, glossy surfaces, concept art, illustration, painting, anime, game environment, dramatic glow, flashlight beam, fluorescent lights, clean surfaces, gore, blood, text, watermark

    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  10. 10

    Generate a reference image 4

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Generate a reference image prompt
    The camera creeps laterally past the edge of the lower bathroom closed door, using the narrow gap to spy inside. The movement is slow and suspenseful, with the dirty door frame crossing close in front of the lens. Inside is only an empty decayed toilet stall: stained porcelain, cracked tile walls, damp grime, mold growth, and a wet reflective floor. Nothing is inside, but the emptiness feels deeply wrong. Photoreal, dark cinematic horror, realistic lighting, heavy atmosphere, slow tension-building camera slide. the camera is from the floor pointed up. you barely can see the toe floating behind the door, after looking through the bottom door gap

    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  11. 11

    Generate a reference image 5

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Image Model: NanoBanana Pro
    2. give me a further view of this, wider shot, a lot further. leg is still visible. super dark, no light, night time
    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  12. 12

    Generate a reference image 6

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Generate a reference image prompt
    Raw low-light photo of a real long abandoned hospital corridor, documentary realism, very dark, heavily underexposed, no visible light source, not cinematic, not CGI. A real condemned hospital walkway photographed in near total darkness. The image should feel like an actual raw low-light photo taken by someone standing alone in a long corridor inside an abandoned hospital. Natural slightly imperfect framing, realistic low-light noise, crushed shadows, limited dynamic range, soft detail loss in dark areas, and severe underexposure. No phone, hand, flashlight, or person visible in frame. The corridor is extremely long, narrow, and deeply unsettling. Dirty cracked tile floor stretching far into the distance, stained and peeling walls on both sides, heavy water damage, black mold creeping along corners and ceiling edges, broken ceiling panels, exposed pipes, damaged wall fixtures, and old hospital room doors lining the hallway. Some doors are closed, some slightly open, some hanging damaged or crooked. The hallway should feel damp, neglected, silent, and airless. Scattered debris across the floor, including broken plaster, dust, paper scraps, rust stains, dirt, and abandoned equipment fragments. The far end of the corridor should disappear completely into darkness. There should be no visible light source anywhere. No fluorescent lights, no moonlight, no doorway light spill, no bright highlights, no dramatic horror glow. The scene should be almost entirely swallowed by darkness, with only faint shapes and surfaces barely readable because of the camera sensor trying to capture the scene. Let the darkness dominate the image. The atmosphere should feel oppressive, claustrophobic, and real. Materials must look physically real: grimy tile, wet patches on the floor, stained plaster, rusted metal, moldy surfaces, peeling paint, damaged doors, and realistic decay. The final result should look like a real severely underexposed photo of a hospital corridor taken in almost complete darkness. Realism cues: raw low-light photo documentary realism very dark heavily underexposed crushed shadows natural image noise limited dynamic range soft shadow detail urban exploration photo amateur realism found footage feel no visible light source Negative prompt: CGI, 3D render, Unreal Engine, cinematic lighting, movie still, stylized horror, glossy surfaces, perfect symmetry, concept art, illustration, painting, anime, game environment, flashlight beam, fluorescent lights, visible window glow, monsters, gore, blood, text, watermark

    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  13. 13

    Generate the video

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Now bring your frame to life in Seedance 2.0.
    2. I am using Higgsfield for Seedance 2.0, but you can use any other platform that offers Seedance 2.0 to achieve the same result.
    3. Open Higgsfield → click Video → Select Seedance 2.0 1
    4. Upload all the assets needed for your scenes, then write a prompt describing them.
    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a reference image.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  14. 14

    Generate a video clip 1

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Generate a video clip prompt
    The black-haired man from the reference image and his cat
    
    15-second vertical video, realistic iPhone front-facing selfie camera vlog quality, handheld at arm's length, slightly amateur framing, recording device never visible in frame, no phone visible, no selfie stick visible, natural hand shake, soft autofocus breathing, low-light phone grain, uneven exposure, slight motion blur, casual off-center selfie composition, not cinematic, not polished, real-life look. A realistic Asian man is doing a ghost hunting vlog inside an abandoned hospital at night. He is exploring a dark, decayed hospital entrance and corridor with peeling walls, broken ceiling panels, cracked floors, rusted doors, abandoned wheelchair, debris, and deep shadows. The place feels wet, silent, creepy, and haunted. He is whispering most of the time because he is scared but trying to act brave. He is holding a flashlight off-screen in one hand, and the flashlight beam is the only source of light in the entire place. The beam shakes naturally as he walks and points around, creating harsh moving shadows and uneven lighting across the corridor. Exactly one single ragdoll cat only, no extra cats, no duplicate cats, no other animals. The cat is clearly with him in the vlog and meows once, shocking him. Scene 1, 1.0 to 5.0. Front-facing selfie-vlog perspective. He stands just outside or just inside the abandoned hospital entrance, filming himself from the off-screen front camera while the flashlight beam lights his face and the broken doorway behind him. His framing is slightly messy and amateur, like a real nervous phone vlog. He whispers to the camera, "Today I'm going into this abandoned hospital to look for ghosts..." Then he slowly starts walking inside, the flashlight beam wobbling across the walls, broken glass, rusted frames, and dark corridor ahead. Exactly one ragdoll cat is with him, close by and visible briefly near his side or being held close to him, but only one cat. Scene 2, 6.0 to 10.0. Still in the same front-facing handheld selfie shot, he walks deeper into the corridor while whispering and scanning the hallway with the flashlight. The beam reveals peeling paint, hanging ceiling panels, scattered papers, cracked floor tiles, and an abandoned wheelchair. The corridor is extremely dark except for the flashlight. He suddenly pauses because he thinks he heard something in the distance. He looks off-camera into the darkness, breathing a little faster, then whispers, "Wait... I heard something..." He tries to laugh it off and says quietly but nervously, "I'm not afraid of no ghost..." Scene 3, 11.0 to 15.0. He keeps moving forward in the same front-facing selfie-vlog perspective, still lit only by the shaking flashlight. The hospital corridor behind him is dark and threatening, with the beam catching doorways and corners for brief moments. Then exactly one ragdoll cat suddenly meows nearby. The sound shocks him and he flinches hard, jerking the flashlight beam and camera slightly. He looks startled, then realizes it is his cat and whispers sharply, "Shiii... don't do that..." He looks back into the camera, still nervous, and keeps walking as the flashlight shakes over the ruined hospital walls and darkness behind him. The cat is exactly one single ragdoll cat only, fluffy long-haired coat, cream-white fur, soft gray points on the ears and face, bright blue eyes, pink nose, and large plume-like tail. No extra cats anywhere in the video. The entire video must feel like a real front-facing iPhone selfie vlog recorded in a genuinely dark abandoned hospital, with the flashlight as the only light source, nervous whispering, realistic fear, creepy silence, and one sudden cat meow scare.

    Expected result for generate the video.
    Use this source result as the visual comparison point for the step.
    Expected result for generate the video.
    Use this source result as the visual comparison point for the step.
    Expected result for generate the video.
    Use this source result as the visual comparison point for the step.
    Expected result for generate the video.
    Use this source result as the visual comparison point for the step.
    Expected result for generate the video.
    Use this source result as the visual comparison point for the step.
    Expected result for generate the video.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  15. 15

    Generate a video clip 2

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Generate a video clip prompt
    The black-haired man from the reference image and his cat
    
    15-second vertical video, realistic iPhone front-facing selfie camera vlog quality, handheld at arm's length, slightly amateur framing, recording device never visible in frame, no phone visible, no selfie stick visible, natural hand shake, soft autofocus breathing, low-light phone grain, uneven exposure, slight motion blur, casual off-center selfie composition, not cinematic, not polished, real-life look.
    
    Scene 1, 1.0 to 4.0 He is walks carefully into the abandoned toilet area Then he suddenly freezes in shock when the flashlight beam catches a pair of pale human legs visible from a distance under a dirty toilet stall door far down the corridor. He panics, breathes hard, and whispers in fear, "What the hell is that...?" He pans the camera angle shot angling the camera enough to clearly show the stall with the legs in the background. Scene 2, 5.0 to 10.0 Still in front-facing selfie camera mode, he crouches lower and almost crawls down to get a lower angle on the stall while keeping his own terrified face partly in frame. He stays several feet away from the stall and does not go near it. He nervously whispers, "Okay... okay... I got a good idea..." From a clear distance, he wispers "hello kitty kitty, come come." then he throws a cat treat across the wet floor directly toward the gap under the stall door. The single ragdoll cat runs all the way in toward the stall, chasing the treat and going right up into the dark stall entrance area. Make this very clear: the cat really goes in toward the door gap and reaches the stall. Right as the cat gets in, the pale legs unnaturally float upward and disappear behind the toilet stall door, as if whatever is there is lifting off the ground. Scene 3, 11.0 to 15.0 The moment he sees the legs float up, he panics instantly and runs out in fear, but the shot must remain front-facing selfie camera the entire time. No external camera angle. No third-person view. He is still holding the selfie camera while running, so his terrified face stays partly in frame as the corridor shakes and swings behind him. The flashlight beam jerks wildly across the walls and floor as he sprints away in vlog style. The whole escape must feel like true iPhone front-facing selfie footage from a terrified ghost-hunting vlogger running out of a haunted hospital toilet while still filming himself. Hard constraints: The cat treat must be thrown from a distance. The cat must clearly run in toward the stall and reach the door gap area. The legs must float upward only after the cat goes in. The running section must stay front-facing selfie camera only. No outside camera angle during the escape. No cinematic chase shot. No third-person running shot. no subtitle

    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  16. 16

    Generate a video clip 3

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Copy the prompt below and replace any project-specific subject, product, location, or format details.
    2. Attach the source assets in the same order described by the prompt, then run one controlled test.
    Working asset
    Generate a video clip prompt
    The black-haired man from the reference image and his cat
    
    6-second vertical video. Real iPhone front-facing selfie camera vlog. The iPhone selfie camera is the main and only camera angle for the entire video. The viewer is the phone. The man is filming himself at arm's length while walking. His face stays in frame most of the time. The shot must look like true handheld selfie footage from a nervous vlogger, not a cinematic horror film. A realistic Asian man is inside a dark abandoned hospital, tired, scared, out of breath, and whispering while walking forward down the hallway. He is vlogging directly to his iPhone front camera the whole time. He is holding a black torch light in his free hand, and the black torch is the only light source in the entire scene. The flashlight beam is narrow, shaky, harsh, and uneven, lighting only parts of his face, the nearby walls, and the path ahead. The rest of the hospital falls into darkness. The framing must feel amateur and natural. Slightly off-center face. Sometimes too close to the lens. Slight bobbing from footsteps. Small hand tremors. Uneven exposure. Soft autofocus breathing. Low-light iPhone grain. Slight motion blur. Imperfect handheld stabilization. Real vlog energy. Not polished. He walks straight through a ruined hospital corridor with peeling walls, wet cracked floor tiles, rust stains, broken ceiling panels, open doors, dark rooms, and oppressive silence. No cat visible anywhere in this video. The cat is missing and never appears on screen. Clip 1. He walks forward in selfie mode, breathing hard, glancing behind him and into side rooms while whispering directly to the camera, "I lost my cat... I don't know where he went..." The black torch shakes in his other hand and throws messy light across his cheek, the corridor wall, and the floor. His face stays dominant in frame while the hallway drifts behind him. Hard cut. Clip 2. New selfie-vlog moment deeper in the hospital. He is still walking forward, more panicked now, still whispering straight into the iPhone front camera. The black torch is still the only source of light. The beam flickers across a darker hallway junction, ruined doorway, and black rooms behind him. He whispers, "I hope he's okay... I gotta get out first..." He keeps moving forward like a real scared vlogger trying to escape. Hard rules: front-facing iPhone selfie camera only handheld selfie vlog only face visible in frame most of the time walking while talking to camera recording device never visible in frame black torch in free hand torch is the only source of light 2 separate vlog clips with a hard cut real amateur handheld vlog feel not cinematic not stabilized not smooth not third-person not security camera not found footage from a distance not over-the-shoulder not dolly shot not gimbal shot no cat visible

    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Expected result for generate a video clip.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  17. 17

    Set the edit rhythm

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Raw AI footage is your base. Post-production is where it becomes cinematic.
    2. Temporal Remapping (Speed Ramps):
    3. Use speed curves to create rhythm and emphasize key moments.
    Source example for generate a video clip.
    Use this visual to confirm the expected setup or result.
    Source example for generate a video clip.
    Use this visual to confirm the expected setup or result.
    Source example for generate a video clip.
    Use this visual to confirm the expected setup or result.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  18. 18

    Grade the colour

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Visual Enhancement: Glow
    2. Apply a "Soft Glow" or "Edge Glow" effect to simulate high-end lens diffusion and luxury aesthetics.
    3. Color Grading: The Final Look
    4. Final color adjustments allow you to refine the mood beyond the AI's base
    Expected result for set the edit rhythm.
    Use this source result as the visual comparison point for the step.
    Expected result for set the edit rhythm.
    Use this source result as the visual comparison point for the step.
    Expected result for set the edit rhythm.
    Use this source result as the visual comparison point for the step.
    Expected result for set the edit rhythm.
    Use this source result as the visual comparison point for the step.
    Expected result for set the edit rhythm.
    Use this source result as the visual comparison point for the step.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.
  19. 19

    Polish and export the final edit

    Use the source inputs to complete this stage and produce a result that can be checked before continuing.

    1. Polish is what makes it feel expensive. Fine-tuning is what maximizes impact and gives your commercial a premium finish.
    Source example for grade the colour.
    Use this visual to confirm the expected setup or result.
    Common failure modes
    • The generated scene changes the person, camera, room, product, or timing beyond the requested effect.
    • Fast motion hides face, hand, clothing, text, or continuity errors that become visible when the clip is paused.

Before you use it

Final checks

  • Check the full result for changed people, products, text, timing, or layout.
  • Confirm the tool, model, controls, and cost before generating.
  • Keep private or unreleased material out of third-party tools.

Share your work

Made it? Show me.

Post your finished result on Instagram and tag @terencesia. I'll share a few standout results with the community.

Tag @terencesia on Instagram (opens in a new tab)

Next step

Run one low-cost test, record the first visible failure, and revise only the instruction or source asset tied to that problem.

Back to the tutorial library

Sources

What this guide relies on