Liminal horror
Backrooms AI Video Generator
A restrained found-footage look for Halloween. The still establishes the corridor and the distant figure; the motion prompt adds a slow walk, a light flicker, and a quiet exit.
- Original liminal-space scene, not a game copy
- Found-footage camera sway and fluorescent flicker
- Eerie without gore or a jump scare
Updated October 20, 2026
Related Halloween looks
Eight original looks built around this year's trending costume language: gothic braids, celestial princess, sci-fi visor hero, pumpkin pet, robot, haunted mascot, liminal horror, and a cowboy-angel couple.
How it works
1
Upload one clear photo
Use a front-facing photo with even light and no sunglasses. One person works best; couples need a multi-reference model.
2
Keep the identity lines
The prompt protects the face, skin tone, body proportions, and the number of people. Change the costume and scene, not the person.
3
Edit once, then animate
Generate the still first. When the face and costume look right, send that image to an image-to-video model for a five-second clip.
Prompt recipe
Generate the still at a wide aspect ratio. The corridor needs horizontal room for the camera move to read as found footage.
Open the image generator, upload your photo, and keep the identity-preservation lines intact. The prompt is written for an image-editing model, not a generic text-to-image prompt.
Cinematic still from an original liminal-space Halloween video: a long empty office corridor with damp beige carpet, stained ceiling tiles, buzzing fluorescent lights, and a distant figure in a simple dark costume standing at the far end. The figure is small and ambiguous, not gory. Handheld camera perspective, slight motion blur, realistic fluorescent color cast, film grain. No text, no logos, no watermark. Do not reproduce a specific film or game scene.
Model routing
Each look can run on more than one model. Start with an identity-first model, then compare a scene-first edit when you need a stronger costume or background.
KIENano Banana ProFace and identity preservation
Image editKIESeedream 4.5 EditCostume, fabric and scene changes
Image editKIEFLUX KontextCostume, fabric and scene changes
Image editKIEGPT-4o Image EditCostume, fabric and scene changes
Image editKIEKling 2.1 ProShort cinematic motion
Image to videoKIEVeo 3.1 FastShort cinematic motion
Image to videoKIESeedance 1.5 ProShort cinematic motion
Image to videoKIESora 2 CharactersShort cinematic motion
Image to videoQuestions
Do I need a photo for this look?+
No. Start in text-to-image, generate the corridor still, then send that still to image-to-video. A reference photo is optional if you want a specific person in the scene.
How do I keep the corridor geometry stable?+
Use a short clip and a simple camera move. Long walks and fast turns give the model more room to invent new doors, lights, or wall angles.
Can I add a person?+
Yes. Keep the figure distant and small; close-up faces make the shot feel like a different genre. If you use a reference photo, say that the figure must stay at the far end of the corridor.
Backrooms AI Video Generator
Generate the still first. When the face and costume look right, send that image to an image-to-video model for a five-second clip.
Original AI interpretations. Vidle is not affiliated with film, game, or costume rights holders. Only upload photos of people who have given you permission.