Halloween AI Trend: Ghost Selfie Step-by-Step + Prompts

👁 126 reads

The Halloween AI trend creates a 2000s-style selfie taken in a dark room with camera flash, showing a figure behind you in the doorway: a bedsheet ghost, someone in a mask, or a pumpkin-headed character. You can make it in Gemini (Imagen 2) or ChatGPT from a single regular photo. The trend exploded in fall 2025 and returned in October 2026 with new versions. We tested two variants and share 8 prompts below. For all current trends, see our AI photo trends roundup.

Where the Doorway Figure Trend Came From

The first version of the trend appeared on TikTok in late September 2025: a Y2K-style girl lying on pink satin bedding with a corded phone, while a figure in a Ghostface mask from the “Scream” series stands in the doorway of a dark hallway. The Tab on October 1, 2025 counted this prompt among the most popular Gemini prompts on TikTok, and Perfect Corp described how the trend combined Halloween season with nostalgia for 90s bedrooms.

Google in its Halloween post from October 2025 showed the same frame, but with a ghost in a white bedsheet instead of the movie character. In 2026, this version became the standard: Gaurav Tiwari’s guide from October 3, 2026 calls the bedroom doorway ghost the biggest Halloween AI trend of the previous year and swaps the movie character for a bedsheet, because apps often reject trademarked characters.

If you’re looking for costume or makeup ideas rather than a spooky frame, check out our articles Halloween costume in AI and Halloween makeup in AI.

How to Create a Halloween AI Selfie Step-by-Step

1. Choose a face photo taken straight-on, ideally from the waist up. A smile works better than a serious expression, because the contrast with the figure in the background creates the whole effect.
2. In Gemini, select image generation; in ChatGPT, tap “+” and add the photo.
3. Paste prompt 1. Describe the figure by appearance (“tall figure in a white bedsheet”), not by movie name.
4. If the model adds posters with real people or logos, write in the same conversation: “replace posters with made-up ones, no real people.”
5. Want a video version? Create the image, then animate it in a video generator (instructions below).

Our Test: Ghost in Doorway and Pumpkin in Mirror

The woman in the examples is AI-generated and does not exist. We first generated a regular “phone photo,” then transformed it in Imagen 2 through OpenRouter using prompts 1 and 2 from the list below.

What we noticed:

• The face, hairstyle and mustard sweater transferred unchanged, and the model automatically transformed the cafe into a bedroom with an outstretched arm as in a selfie.
• The date “31 10 2004” came out exactly as we typed it, in orange digits like old cameras.
• In the first version (without the poster note), the wall had posters of real bands and a movie with recognizable faces. We’re not publishing that image. The addition “made-up bands, no real people” solved it.
• In the mirror version, the model wrote “06/10/26” in the corner instead of the year 2006. But the pumpkin-headed figure in the reflection looks exactly as we wanted.

8 Halloween AI Prompts

We tested prompts 1 and 2 ourselves; the rest follow the same structure but we haven’t tested them. Insert your own details in square brackets.

Prompt to copy (1, bedroom doorway ghost, tested):

Turn the attached photo into a 2004 bedroom selfie taken with an old digital camera with direct flash. Same person sitting on the bed holding the camera at arm's length, slightly overexposed face, dark room, posters of made-up bands on the wall with no real people and no known logos, orange date stamp "31 10 2004" in the corner. In the open doorway behind, in a dark hallway, a tall figure in a white bedsheet with cut-out eye holes stands still. The person doesn't notice it and smiles at the camera. Keep the face and hair exactly the same.

What it’s for: main version of the trend. Watch out: we tested it in Polish; adjust pronouns as needed for your photo.

Prompt to copy (2, pumpkin in mirror, tested):

Turn the attached photo into a 2006 bathroom mirror selfie. Same person holding a flip phone, flash reflecting in the mirror, greenish fluorescent light, period tiles. In the mirror, behind their shoulder, in a dark hallway stands a figure with a carved pumpkin instead of a head, with glowing eyes. The person sees nothing. Keep the face, hair and earrings exactly the same. Grain, low quality like a phone from 20 years ago.

Watch out: if you want a date, write it in full (“date 31.10.2006”), because the model invented its own for us.

Prompt to copy (3, corded phone, 2025 version without movie character):

Create a photo of the person from the attached photo in Y2K style: lying on shiny purple satin bedding talking on a large corded phone from the 90s, next to a bowl of popcorn and old magazines. Room lit only by a small lamp. In the doorway of the dark hallway behind them stands someone in a black cloak and white mask, staring motionless. Grain like a 90s camera. Keep the face from the original.

Prompt to copy (4, couple under bedsheets):

Turn the two people from the attached photo into classic bedsheet ghosts with cut-out eye holes. Keep their height, positions and background from the photo. They're standing on an apartment building staircase, lit by flash, like a party photo from 2003.

What it’s for: shared Halloween photo without showing faces.

Prompt to copy (5, disposable camera photo):

Turn the attached photo into a disposable camera photo from a 2003 Halloween party: same person in [black witch hat] holding a cup of punch and laughing, flash, red eyes, pumpkins with candles in background. Behind the window behind their back, in the darkness, a pale expressionless face is barely visible. Keep the face from the original.

Prompt to copy (6, bedsheet ghost):

Turn the attached photo into a 2004 bedroom selfie taken with an old digital camera with direct flash. Same person sitting on the bed holding the camera at arm's length, slightly overexposed face, dark room, posters of made-up bands on the wall with no real people, orange date stamp "31 10 2004" in the corner. In the open doorway behind, in a dark hallway, a tall figure in a white bedsheet ghost costume with cut-out eye holes stands still. The person doesn't notice it and smiles. Keep the face and hair exactly the same.

Prompt to copy (7, Victorian ghost, version from Google post):

Transform this person into a semi-skeletal, translucent Victorian ghost in Victorian clothing. Maintain the existing pose and framing. Ultra-realistic, black-and-white daguerreotype, wet plate photography with streaks and stains.

Prompt to copy (8, video from image):

The person keeps smiling at the camera while the ghost in the doorway behind slowly tilts its head and takes one step forward. Camera flash flickers once. Static handheld shot, grainy 2000s video look, no music.

What it’s for: second step in a video generator, with the image from prompt 1 as the first frame.

Which Tools and Is It Free

The Gemini help page (as of October 3, 2026) confirms that Imagen 2 works without a plan, but no longer lists the number of images. From October 9, the free Gemini chat switches to Gemini 2.0 Flash Lite, and Google wrote nothing about images. How to animate a finished image is shown in our articles Hailuo 3.0: image to video and Kling AI.

What to Watch Out For

• Movie characters (“Scream,” “Saw”): apps often block them. Describe the costume, not the name.
• Posters and shirts with real people in the background: check every photo before publishing, because the model can draw them without asking.
• Blood and wounds in the prompt: common reason for rejection. The atmosphere comes from darkness and the figure in the background, not effects.
• Photos of children: don’t transform them into scary scenes. Other people’s photos only with permission, and mark the finished image as AI.

Frequently Asked Questions

How do you make the Halloween AI trend for free?

Upload your photo to free Gemini or ChatGPT and paste prompt 1. It’s one command, no additional apps.

Why won’t Gemini add Ghostface?

It’s a trademarked character. Describe it by appearance (“figure in black cloak and white mask”) or swap it for a bedsheet ghost, like Google did in their post.

How do you turn this photo into a video?

Upload the finished image as the first frame to a video generator (Kling, Hailuo, Higgsfield) and describe the movement, for example with prompt 8.

How is this different from Halloween costume in AI?

A costume changes your outfit, while this trend leaves you unchanged and adds a scary figure in the background. They’re two separate photo ideas.

Kinetyra

The link to Higgsfield is a partner link: if you buy through it, QUANTSLANT earns a commission and you pay the same price. It has no influence on the content, including where the coverage is critical.