Halloween AI Prompts

Each card contains a complete prompt, not a keyword list. Paste it into the image-to-image Studio after uploading a photo, or into text-to-image if you want the model to invent the person. The prompts are intentionally explicit about identity, anatomy and negative outcomes because those are the three places Halloween images usually fail.

Change one variable at a time. If you want a different colour, change the colour. If you want a different room, change the room. If you change the costume, the hair, the background and the lighting all at once, you will not know which instruction caused the problem. One variable per attempt is slower to write and faster to finish.

Every reliable prompt on this page follows the same order: preserve, change, place, light, avoid. Preserve names the face, skin tone, pose and number of people. Change names the costume, hair or makeup. Place names the room, street or landscape. Light names the direction, colour and softness. Avoid names the failure modes you do not want.

This order also makes the prompt easier to edit. If the face drifts, strengthen the preserve line. If the costume is wrong, add fabric and silhouette details. If the image looks flat, add a specific light source. If the model adds extra fingers or people, repeat the avoid line and reduce the amount of action in the scene.

Single portraits are the most reliable input. Use a sharp photo with even light and keep the framing close to the original. The Selfie Sorceress prompt adds a velvet coat, silver jewellery and a candlelit library. The Skeleton Glam prompt changes makeup only, which is the lowest-risk way to make a portrait feel Halloween without losing the person.

If you want a mask or a full character head, say “wearable costume” and “keep the human face visible.” Otherwise the model may replace the face entirely. That can be a valid artistic choice, but it is a different image and usually a different use case.

Couple prompts need two identity protections. Name both people and say “do not add a third person.” Keep the pose simple: standing together, shoulders visible, both faces unobstructed. A matching palette helps the image feel intentional: deep plum, antique gold and cool moonlight work well for a Halloween couple.

For groups larger than two, generate each person separately with the same background and lighting instructions, then combine the approved images. A single group generation is possible, but every additional face increases the chance that one identity drifts or one person is duplicated.

Pet prompts must protect species, breed, fur pattern and anatomy. Say “keep the animal’s original breed, fur colour, markings, eye colour and body proportions.” Then describe the costume as something worn, not something that replaces the body. Avoid costumes that cover the eyes or change the shape of the snout.

A front-facing pet photo with visible eyes and ears works best. If the photo is blurry or the pet is moving, the costume edges will be soft. Use a simple studio or autumn set so the model spends its attention on the animal rather than on ten background props.

A video prompt needs one motion, not a storyboard. Start from an approved still, repeat what must stay the same, then describe one action and one atmospheric movement: “turns her head slowly; fog drifts; the camera pushes in gently.” Add “no scene cuts, no identity change, no extra people.”

Five seconds is enough for a head turn, a lantern flicker or a short push-in. If you need a longer sequence, generate separate shots and edit them together. Trying to fit a walk, a turn and a spell into one five-second clip is the fastest way to get warped anatomy.

The best prompt is the one that describes your photo. If you are wearing glasses, say whether they should stay. If you have curly hair, say “keep the curl pattern.” If the photo is backlit, ask the model to preserve the rim light. Those details are more useful than another adjective.

Keep a small notebook of prompts that worked. Change the date, the colour palette and the setting for next year, and you have a reusable seasonal workflow instead of a one-off experiment. The same structure works for Christmas, New Year and Valentine’s Day.