15 prompts for Halloween selfies, costumes, pets, couples and image-to-video. Each one states what to keep, what to change and what to avoid, so it works in Gemini, ChatGPT or any image editor that accepts a reference photo.
Change one thing per attempt. Stacking costume, background and pose changes in one prompt is the most common reason a result looks wrong.
Halloween selfie prompts
Keep the face and pose untouched, then change only the styling. Works with any image editor that accepts a reference photo.
Haunted unicorn selfie
Dress the subject as a haunted unicorn mascot: soft plush hood with one spiraled horn, matted mane, stitched velvet panels and slightly eerie eyes. Moonlit pumpkin field with hanging lanterns, cold blue rim light against warm orange fill. Cinematic editorial photograph, visible plush and fabric texture. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Restyle the subject with long purple braids and small silver hair ornaments, in black and purple velvet gothic clothing with lace and metal details. Moonlit Victorian garden, thin fog, lanterns, dark autumn leaves. Realistic magazine photography, cinematic soft light, natural skin texture. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Dress the subject in a midnight-blue fantasy gown with silver embroidery and softly glowing star patterns on flowing fabric, plus a silver moon crown and glowing accessories. Moonlit observatory with tall windows onto a star field. Elegant fantasy fashion photography, cinematic rim light, deep blue and silver palette. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Keep the photo exactly as it is and replace only the makeup: pale base, soft grey contour around the eyes, thin hand-drawn bone lines on the cheeks and a muted berry lip. Do not change the hairstyle, clothing, background or lighting. Realistic skin texture, no sticker effects. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
The safest first test: one variable at a time.
Costume and scene prompts
Full costume changes need a stronger scene description so the model has somewhere to put the character.
Galactic hero portrait
Dress the subject in a wearable sci-fi hero costume: armored chest plates, joint guards, matte fabric underlayer and blue LED accents, with a translucent visor pushed up above the eyes. Blue and orange cinematic studio lighting, believable material reflections. Photograph a person wearing a costume, not a robot character. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Dress the subject in a black velvet masquerade outfit with a feathered half mask held in one hand, and place them in a candlelit stone hall with tall mirrors. Keep the face fully visible and unchanged. Cinematic photography, warm candle falloff, deep shadows, no text or watermark. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Victorian ghost portrait
Dress the subject in a faded Victorian high-collar dress or coat with a lace veil, standing in a foggy graveyard at dusk. Keep the face recognizable and the skin natural rather than translucent white. Desaturated palette, soft overcast light, film grain, no blood and no text. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Pet costume prompts
Pet prompts fail in one specific way: the costume hides the animal’s face. Always state that the eyes and muzzle stay clear.
Pumpkin pet
Dress the pet in a soft fabric pumpkin costume with a small green leaf collar, fitted so the eyes and muzzle stay clear. Warm autumn studio backdrop with small pumpkins and dry leaves. Professional pet portrait photography, realistic fur texture, gentle catchlights in the eyes. Preserve the animal's breed, coat colour, markings, eye colour and facial structure. Keep the same species and number of animals. Keep the eyes, muzzle and ears clear. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Dress the pet in a soft felt dinosaur costume with a row of rounded back spikes that faintly glow green, fitted so the eyes, muzzle and ears stay visible. Dark studio backdrop with a single soft key light. Realistic fur texture, the pet stays the same breed and colour, no extra animals. Preserve the animal's breed, coat colour, markings, eye colour and facial structure. Keep the same species and number of animals. Keep the eyes, muzzle and ears clear. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Butter stick
Dress the pet in a simple rectangular butter-stick costume in pale yellow fabric with a plain wrapper-style band, light enough that the pet can still walk. Clean kitchen counter backdrop, warm daylight. Keep the pet’s markings, eye colour and body shape unchanged, no brand logos or text. Preserve the animal's breed, coat colour, markings, eye colour and facial structure. Keep the same species and number of animals. Keep the eyes, muzzle and ears clear. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Couple prompts
Two people in one frame needs a model that accepts two reference images. Upload both portraits and name who wears what.
Cowboy and angel
Create one coordinated photo of two people: the first person in a western cowboy outfit with a hat and fringed jacket, the second in a soft white angel outfit with feathered wings. Keep both faces, hairstyles and body proportions unchanged, and keep exactly two people in frame. Warm dusk light on a dusty open road, cinematic photography. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Requires a two-image reference workflow.
Hero and pilot
Create one coordinated photo of two people: the first in an armored hero suit with a visor pushed up, the second in a worn flight suit with a helmet under one arm. Keep both faces recognizable and the same height difference as the references. Hangar backdrop with cool practical lights, cinematic colour grading, exactly two people, no text. Preserve the person's recognizable face, facial proportions, skin tone, age and body proportions. Keep the same number of people, the original framing and pose. Change hair only when the selected look requests it. Do not swap faces, add extra limbs or subjects, duplicate fingers, distort anatomy, or add text, logos or watermarks.
Image-to-video prompts
Start from a still you already like, then ask for one small movement. Blinks, a head turn and drifting fog survive compression far better than choreography.
Slow head turn with drifting fog
Use the provided image as the exact first frame and visual reference. Five seconds of continuous cinematic portrait motion: the subject slowly turns their head toward the camera with a calm, unreadable expression while fog drifts through the background and a few strands of hair move. Keep the face, costume, hairstyle and body proportions identical throughout. Slow, steady camera push-in, stable structure, consistent lighting, no cuts, no new characters, no text or watermark.
Candle flicker and glowing detail
Use the provided image as the exact first frame. Five seconds of subtle motion: candle flames flicker, the glowing details on the costume pulse gently, and the subject blinks once. The camera stays nearly still with a very slow drift. Keep identity, costume and framing unchanged, no cuts, no added characters, no text.
Lantern rise in the background
Use the provided image as the exact first frame. Five seconds: several paper lanterns rise slowly behind the subject while leaves drift past in the foreground and the subject holds still with one small breath. Keep the subject’s face and costume identical, keep the camera locked, no cuts, no new people, no text or watermark.
Want a ready-made look instead?
The costume generator offers the five prepared looks and asks to preserve your face, and returns an image you can download.
The prompt is only half of the result. The reference photo and the model you use decide the rest.
Start from a photo, not from text
Paste the prompt into an image editor and attach a reference photo. Text-only generation will invent a new person instead of dressing you.
Keep the identity paragraph
Every prompt here already describes what must stay the same. If you shorten it, keep the identity and exclusion sentences and cut the scene instead.
Iterate, then animate
Once a still works, use it as the first frame of an image-to-video prompt. Short movements such as a head turn or drifting fog survive compression far better than choreography.
Halloween Prompt FAQ
Do these prompts work in Gemini and ChatGPT?
They are written as plain instructions, so they work in any assistant or image editor that accepts an uploaded photo and edits it. Expect different levels of identity preservation between models.
Why do my results add extra people or fingers?
Long prompts that describe a whole scene give the model more room to invent. Keep one subject, state the number of people, and rerun rather than adding more instructions.
Can I use a prompt commercially?
Using a prompt is fine. The rights to the output depend on your rights to the input photo and on the model provider’s terms, and a paid plan does not grant rights to third-party characters or likenesses.
What should I do for a couple photo?
Use a workflow that accepts two reference images, one per person, and state which person wears which costume. Single-image models tend to merge two faces into one.
