How to make an AI Halloween transformation video

Turn this person into an elegant vampire: pale skin, dark red lips, subtle fangs, slicked-back hair, a high-collared black cape with a deep red lining. Keep their face, expression, pose and the framing exactly the same. Change the background to a candlelit stone hallway. Photorealistic, soft warm candlelight.

Why it works: It says what changes and, just as clearly, what must stay, so the person is still recognisably themselves.

Dress this person as a scarecrow: a patched flannel shirt, straw sticking out of the cuffs and collar, a battered wide-brimmed hat, stitched-on face paint around the mouth. Keep their face, pose and framing the same. Put them in a pumpkin patch at dusk with a low orange sky and a thin ground fog.

Why it works: Face paint and props read as a costume while the face underneath stays the same person.

@me-before stands in a bright living room and looks into the camera. The camera whips fast to the right with heavy motion blur and lands on @me-halloween in the same pose, same spot in frame, now in a dim candlelit room. They slowly smile. Vertical, handheld phone feel.

Why it works: The motion blur hides the cut, and keeping the same pose and spot in frame makes it read as one person changing.

@me-before looks into a bathroom mirror and the lights flicker. In the reflection, @me-halloween looks back. The person turns from the mirror toward the camera and is now @me-halloween too. Slow push-in, cold blue light with one warm flicker, quiet hum, then silence.

Why it works: A mirror gives the change a reason to happen on screen, and the turn to camera lands the reveal.

A figure under a plain white bedsheet with two cut-out eye holes stands in a dark hallway. The sheet drops to the floor in one motion, revealing @me-halloween standing in the same place, lit from below by a flashlight. Static camera, eye level, a soft whoosh as the sheet falls.

Why it works: The sheet is a natural wipe: the reveal happens in a single physical action the model can follow.

Put the face from the new-face photo on the person in the black cape. Keep their costume, makeup, moves, the candlelit set and the sound exactly as they are.

Why it works: Video face swap only changes the face, so the costume, moves and set of the clip you generated carry over.

The character in Image 1, in their full skeleton costume with glowing face paint, performs the dance from Video 1, matching every move, the camera and the timing. Put them in a foggy suburban street lit by jack-o'-lanterns.

Why it works: Motion transfer copies the moves of a dance clip, so your costumed character joins a trend without anyone being filmed in costume.

Restyle this photo as a 1990s film still: grainy, warm tungsten bedside lamp, a corded phone in the person's hand, pastel bedding and posters on the wall. In the dark doorway behind them, add a tall masked figure in a black hooded robe, half in shadow, watching. Keep the person's face and pose the same.

Why it works: This is the widely shared masked-intruder look described generically, so it needs no franchise name to work.

@your-character sits on the floor of a sunlit vintage bedroom wearing a white bedsheet ghost costume with two eye holes, sunglasses on over the sheet, holding a vinyl record. 35mm film photo, soft window light, muted autumn colours, vertical.

Why it works: The deadpan ghost-sheet shoot is a seasonal favourite; the props give it a personality the sheet hides.

Make this photo look like a frame from a 1994 home video tape: soft focus, colour bleed, scan lines, a slight tracking wobble at the bottom, an orange date stamp in the corner reading OCT 31 1994. Add a faint pale figure at the end of the hallway behind the person. Keep the person unchanged.

Why it works: Naming the tape artefacts one by one gets a believable VHS look instead of a generic filter.

A woman in a nightgown at the end of a long, dim hotel corridor holds a candle. The corridor stretches behind her as the camera pulls back and zooms in at the same time. Her face slowly changes as she realises she is not alone. Low hum, a door creaks shut somewhere behind her.

Why it works: The push-pull stretch is the classic horror beat; one face and one realisation keep it readable.

The camera glides over a dead lawn toward a crooked Victorian house at night, through a cracked upstairs window, into a dusty nursery lit by moonlight. A rocking chair starts to rock on its own as the camera arrives. Wind outside, then a slow creak.

Why it works: Outside, then through an opening, then a payoff inside: three stages the template already paces.

Shaky handheld camcorder footage at night: a flashlight beam sweeps across bare trees and fog. The beam stops on a figure standing perfectly still far away between the trees, then the figure is gone. Heavy breathing, crunching leaves, a VHS timestamp in the corner.

Why it works: Found footage hides imperfections by design, which makes it a cheap, forgiving format to draft on Fast.

Extreme low angle from the ground in a cornfield at dusk. A tall scarecrow-like creature with a carved pumpkin head, its eyes glowing orange, slowly turns its head down toward the camera. Backlit by a red sky, corn stalks rustling, a low rumble.

Why it works: Looking up makes anything loom, and a single slow head turn is scarier than a lot of movement.

Dress this dog in a cosy orange pumpkin costume with a green stem hat. Keep the dog's breed, fur, face and pose exactly the same. Put it on a porch with carved jack-o'-lanterns and fairy lights at dusk. Photorealistic, warm light.

Why it works: Keeping the breed, fur and face fixed makes it unmistakably your dog, just in costume.

The cat from image 1 sits on a windowsill wearing a small black witch hat, a full moon outside. It slowly turns to the camera and blinks once, and the candle beside it flickers. Static close-up, soft purple and orange light, a quiet purr.

Why it works: One small action and a static camera keep the cat looking like your cat.

Replace the dog in Video 1 with the dog in a dragon costume from Image 1. Keep every move, the camera and the background exactly as they are.

Why it works: Film your pet once doing anything, then swap in the costumed version without dressing them up again.

The candle from image 1 sits on a dark wooden table between small carved pumpkins. Fog rolls across the frame and clears as the camera slowly pushes in, the flame catches and lights the label. Warm amber light, shallow depth of field, no on-screen text.

Why it works: Fog makes a built-in reveal, and pointing at image 1 keeps your real product in the shot.

Overhead view of hands piping white icing spider webs onto round orange cookies on a black slate board, then placing a candy eye on each one. Even soft light, crisp shadows, satisfying and calm.

Why it works: Top-down hand work is easy to follow on a phone screen and shows the product being made.

@your-character holds up a black velvet cape to the camera in a cluttered costume shop, swings it on in one move and strikes a dramatic pose, then grins. "Last-minute costume? Done." Handheld, warm shop lights, vertical.

Why it works: The line and the action land in the first seconds, which is the whole job of a hook.