The performer in a red robe with white water sleeves, the train spread across a pale floor

Afterimage

A dialogue-free film about Vera, a woman in a concrete room, and the Kunqu performer she finds beyond it.

AI short film, 2026

The idea came from my own experience of being overwhelmed by options, and by AI.

Vera lives alone in a concrete room. She presses small ceramic tiles into a glowing glass wall to enter recorded worlds, where she can become anything and go anywhere. Each visit leaves a small mark on her face.

I made the film with the same generative tools it is about. At the end, Vera stops consuming recorded worlds and makes one thing of her own, a hand-marked tile on a shelf.

It grew out of my MA research at London College of Fashion into generative AI in VFX pipelines. Below is the workflow, stage by stage, as I presented it in my process talk.

Script

For the final film I wrote the script, beat map and shot list before generating any shots.

The script has no dialogue and is in two parts, the opening in Vera's room and the theatre. It describes each shot in physical detail, for example Vera's hand reading the ceramic tiles by touch. Every production image was made for a specific shot.

Script page for The Choosing, a scene from the first part of the film

Script page for The Veiled Theatre, from the second part

Six reference frames of a bare concrete room with bedding, tiles on the floor and cables on the walls — Six reference frames for the room, with tiles scattered across the floor as in the script.

Character and world

Vera and the performer each have a character sheet.

I made a character sheet for Vera and one for the performer. Scene references only set the pose, light and environment. If a face drifted away from the sheet, I regenerated the shot instead of keeping it as a variation. I designed each location around one architectural element, a wall of light for the room and staggered layers of scrim for the theatre.

ToolsNano Banana, Midjourney

Vera front on, pale and cropped-haired, cables running from her neck — Vera's front-on reference.

Character sheet for the Kunqu performer, front, side and back in costume — Character sheet for the performer. Her face, make-up and costume are fixed for every shot.

The performer from the character sheet placed into a theatre scene — The performer placed from her character sheet into a scene, with the costume, silhouette, make-up, face and sleeve construction unchanged.

The concrete room, one wall glowing with white light

The theatre, dark scrim layers with a warm glow behind

Selection

I selected frames shot by shot, with notes on every rejection.

I generated hundreds of images for the room and the theatre. Each frame I kept is tied to a beat in the script. When I rejected a frame, I wrote down what was wrong so the next generation could fix it, as in the collar note on the rejected frame. The colour rule was that red appears first in the architecture, then only on the performer, and never on Vera.

ToolsMidjourney, Magnific, Nano Banana

A rejected frame where the collar has closed into a ring, with a note — Rejected, because the collar closed into a ring. My note asked for the collar to stay open at the throat, with the same framing.

A kept frame, Vera reaching out with one hand — Kept, because the reaching gesture fits its beat in the script.

A grid of nine generated theatre frames, with blue mountains, a pale horizon, glass panels, the performer and a red hall — Nine generated frames for the theatre, before selection.

A grid of nine generated frames of the Kunqu performer in red and white robes, mostly on a pale ground — Nine generated frames of the performer, before selection.

Kept frame: Vera in a dark glass corridor, her reflections either side

Kept frame: a small figure in a dark hall, red light inside the walls — Red in the walls, before it moves to the performer.

Image to video

I matched every clip to rehearsal footage of Kunqu.

The movement is based on rehearsal footage of friends who perform Kunqu. In each generated clip I checked whether the posture, wrist movement, weight shift and the delay in the sleeve came through. I set up two routes, LTX 2.5 in ComfyUI and Seedance, and compared the results shot by shot.

ToolsComfyUI, LTX 2.5, Seedance, Depth Anything

Route one: LTX 2.5 Director in ComfyUI
Per clip: Six seconds, three guiding stills, one local prompt each
Sampling: 12 steps, a 2x latent upscale, then 5 more
Control: Depth Anything V2
Route two: Seedance, one motion prompt per shot

Generated sleeve detail, red linen over a sheer lining

A bowed figure in motion, the cloak trailing behind — Frame from the best six-second test, where the posture, timing, weight and fabric delay all came through.

Finishing

Faces, skin and costume edges were repaired shot by shot.

The overall look of each world came together quickly. Most of the time went into fixing faces, skin, costume edges and flicker between frames. Each shot went through face repair in ComfyUI and an upscale in Topaz, and the edit, grade and sound were finished in DaVinci Resolve. Chapter cards divide the film into sections.

ToolsComfyUI, Topaz, DaVinci Resolve

Hardest to control

Face textureCatchlightsSkinCostume edgesColour shifts between shotsTemporal artefacts

Close-up of an eye before the upscale, soft skin and a flat iris — Before the Topaz upscale, left half of the frame.

The same face after the upscale, with freckles, iris detail and catchlights recovered — After, right half. Freckles, iris texture and skin detail are recovered.

Chapter card: a square of red pigment on cream paper

Chapter card: red ink bleeding into white paper

Chapter card: grey grain fading into black

The performer

Look development and performance frames for The Veiled Theatre, the second part of the film, with the performer in red and white robes against a pale ground.

The performer in white face paint folding long white sleeves over a red robe

The performer bending back, face to the sky, in sheer red silk

The performer seen side-on, a long red train behind her

A teal silk ribbon thrown in an arc behind the performer

The performer in white turning as teal silk streams past

The performer standing in a heavy red cloak over pale skirts

The performer turning in a dark, blurred robe, sleeves outstretched

The performer bent back mid-turn, red silk falling around her

The performer seated behind a low table, sleeves spread across it

The performer in red and white seated over folded silk on a table

Close on a red linen cuff over a sheer white lining

The performer facing forward in a red robe, the train spread wide

Selected frames

Stills from the film and from its visual development.

The performer crossing the stage in a long trailing gown, blue mountains behind

The performer in red seen through shattered glass

Five robed figures in silhouette along a pale horizon

Vera at the edge of a white stage, half in shadow

1 / 6

The performer's painted face in close-up, red around the eyes

Vera's hands lifting a white collar at the back of her neck

Afterimage

AI short film, 2026

0:00

Afterimage

Afterimage

Photography