How to do the ghostface AI trend?
Upload a clear selfie to Google Gemini and ask for a grainy Y2K bedroom portrait with a corded phone, plus a masked figure in a dark doorway behind you. Describe the mask, not the movie character — naming violent franchise villains trips content filters. Gemini image editing requires you to be 18+.
Why — the first-principles explanation
The trend works because of a specific capability shift: image models stopped being generators and became editors. Older tools built a picture from scratch, so your face came out as somebody else's face. Current models take your uploaded photo as a reference and preserve identity across a restyled scene — Google calls this character consistency. That's the whole reason the trend spread. It's recognizably you in the creepy hallway, and "kind of looks like you" would never have gone viral.
The second mechanism explains the weird part everyone hits: your prompt gets refused, and you don't know why. Gemini runs content filters against Google's Prohibited Use Policy, and prompts referencing violent horror-franchise killers can trip them. The model isn't confused about what you want. It's declining. The workaround is to describe the visual, not the IP — "a figure in a white elongated ghost mask and black hooded robe, standing in a dim doorway, out of focus" describes the same image without invoking the character. Filters key on names and violent framing much more than on shapes and fabrics.
The third piece is why the results look so good: the aesthetic is doing the work. Grain, harsh flash, low resolution, and a 90s color cast hide the exact flaws AI images are worst at — over-smooth skin, weird texture, uncanny lighting. When an image is supposed to look like a degraded film photo, all of that reads as authenticity instead of error. This is a general rule worth stealing: whenever you want a convincing AI photo, pick a style that's already imperfect. Clean studio lighting exposes the model; a scratchy disposable-camera look protects it.
One thing worth knowing before you post: the safety and rights obligations sit with you. Google's terms warn against violating others' copyright or privacy rights, and Gemini may remove images when its systems detect a possible policy violation. Putting your own face in a horror scene is fine. Putting a classmate's face in one, without asking, is the version that gets people in real trouble.
An example that makes it click
Think of it like a costume photo booth that only needs one Polaroid of you. You hand the booth your picture and a written description of the scene — pink satin bedsheets, a chunky corded phone, someone standing in the doorway. The booth doesn't paste your head onto a body. It repaints the whole scene while keeping your face as the anchor.
And the booth has a bouncer. Say "put the guy from that slasher movie behind me" and the bouncer stops you at the door. Say "a tall figure in a long white ghost mask, black hood, standing still in the shadows" and you walk right in — because you described a shape, not a character. Same photo. Different words.
How to do it
- Open the Google Gemini app and sign in. Image editing requires users to be 18 or older; image generation is 13+.
- Upload one clear, well-lit selfie. Face fully visible, no sunglasses, no heavy shadow — the model anchors your identity to this photo, so a bad input guarantees a bad output.
- Write the scene prompt first, without the villain: 'Create a dreamy Y2K-style portrait of me lying on pink satin bedding, holding a large 90s corded phone in a daydreaming pose, grainy film look, harsh flash, 1998 disposable camera aesthetic.'
- Add the lurker by description, not by name: 'In the dark doorway behind me, out of focus, a tall figure in a long white elongated ghost mask and black hooded robe, standing perfectly still.' Naming a violent movie character is what triggers refusals.
- If it refuses anyway, soften the framing: remove any words implying threat or violence, keep the figure passive ('standing', 'watching'), and push the horror into lighting and composition instead.
- Iterate in the same chat rather than starting over. Ask for changes one at a time — 'make the doorway darker,' 'add more film grain,' 'move the figure further back.' The model keeps your face consistent across edits.
- Fix the last 10% outside the model if needed: if the mask keeps coming out wrong, generate the room scene alone and composite the figure in with any photo editor.
- Before posting, check it's only your own face. Google's terms warn against violating others' copyright or privacy rights, and images may be removed when systems detect a possible violation.
Key facts
- Google Gemini supports uploading an image and asking for edits, and combining multiple uploaded images into a new one.
- Editing images in Gemini Apps is restricted to users 18 and older; generating images requires users to be 13+ (or the applicable age in their country).
- Nano Banana 2 supports character consistency — maintaining the look of a person or character across generated images — which is the feature that keeps your face recognizable in the restyled scene.
- Gemini Apps may remove images when systems detect a possible violation of Google's Terms of Service, including the Prohibited Use Policy.
- Google's guidance explicitly warns users not to violate others' copyright or privacy rights when generating or editing images.
- The trend's grainy Y2K aesthetic conceals typical AI artifacts — over-smoothed skin and uncanny lighting — because degraded film look makes imperfection read as authenticity.
▶ The 60-second explainer (script)
Here's how the Ghostface AI trend actually works, including the part that keeps failing for people. Open the Gemini app and upload one clear selfie. Good lighting, face fully visible — the model anchors your identity to that photo, so garbage in, garbage out. Now the trick: write the scene first, and leave the villain out of it. Ask for a dreamy Y2K portrait, pink satin bedding, a big 90s corded phone, grainy film, harsh flash, 1998 disposable camera look. Then add the figure — but describe the shape, not the character. Say: a tall figure in a long white ghost mask and black hood, standing still in a dark doorway, out of focus. Do not name the slasher from the movie. That's what trips Google's content filters, and it's why your prompt keeps getting refused. It's not confused. It's declining. One more thing worth knowing — the reason these look so convincing is the grain. Film grain and harsh flash hide exactly what AI is worst at: plastic skin and weird lighting. Pick an already-imperfect style and the flaws read as authenticity. Last thing: use your own face. Google's terms warn against violating other people's privacy, and editing requires you to be eighteen.
What authoritative sources say
People also ask
Why does Gemini refuse my Ghostface prompt?
Naming a violent horror-franchise character can trip Google's Prohibited Use Policy filters. Describe the visual instead — a white elongated ghost mask, black hooded robe, standing in a dim doorway — and keep the framing passive rather than threatening.
Do I need to pay for this?
Gemini's free tier includes image generation and editing with usage limits that vary by plan and change over time. As of 2026-07, check the in-app plan details for current limits before subscribing.
Why doesn't the output look like me?
Your reference selfie is probably too dark, too small, or partly obscured. Character consistency anchors to that upload, so use a sharp, evenly lit photo with your full face visible and no sunglasses.
Can I use a friend's photo?
Don't, unless they've agreed. Google's guidance warns against violating others' privacy or copyright rights, and Gemini may remove images that trip policy detection. Putting someone else's face in a horror scene without consent is the version that causes real problems.
Why do these images look more real than most AI photos?
The Y2K aesthetic hides AI's worst artifacts. Heavy grain, harsh flash, and low resolution disguise over-smoothed skin and uncanny lighting, so flaws read as authentic film degradation instead of mistakes.