Create Four Halloween Portrait Looks from One Photo with GPT Image 2.5

One portrait can become four very different Halloween characters without turning the whole post into a costume shopping list. This AI Halloween photoshoot changes wardrobe, lighting and setting while using the same face as the visual thread.

We tested a velvet vampire, a midnight angel, a gothic witch and a ghost bride. Each request starts from the original portrait, so the fourth look does not inherit changes made to the first three.

Reference input	Gothic witch

Separate the face reference from the costume description

Choose a clear adult portrait with visible facial features and hair. Our synthetic reference has brown skin, short natural hair and small earrings. We ask the model to retain those recognizable details while inventing the outfit and scene; this is character styling, not a test of a specific garment product.

Our reference was generated for this demonstration; no real person, pet owner or location is being represented as the subject of a real event. You can use your own suitable photo or another reference you are authorized to edit.

Reproduce the selected result in WeShop

Open WeShop’s GPT Image generator, attach the reference through the image upload control, and select GPT Image 2.5 Sunburst. The source must actually be attached: describing it in words alone would create a different task.

Setting	Tested value
Tool	WeShop GPT Image generator
Model	gpt-image-2.5-sunburst
Input	One attached reference image
Aspect ratio request	4:5
Image size / quality	1K / medium
Quantity	One per prompt
Selected output dimensions	912 × 1152 pixels

The table reports the requested aspect ratio and the returned dimensions separately. The portrait outputs in this set do not have to be an exact mathematical 4:5 crop. These runs were made on October 20, 2026, with one result per prompt; they do not measure repeat-run reliability.

Give every character a different silhouette

A raised velvet collar, wings, a pointed hat and a veil make the four looks readable as a set. The vampire and witch share a dark mood, so color and props help distinguish them. Keep the face visible rather than hiding it behind a mask or heavy fog.

Return to the original for every look

Attach the same source each time. Name the outfit, prop, background and lighting for that one character, then repeat the facial preservation requirements. This reduces dependence on an already stylized result, although it does not guarantee identity consistency.

Copy the tested gothic witch prompt

Edit image 1 into an editorial gothic witch portrait of the same adult woman. Preserve her recognizable face, skin tone, close-cropped hair and gold earrings. Fully clothed in a deep violet Victorian-style high-neck dress and an elegant broad pointed witch hat. A small antique brass lantern held naturally at waist height. Misty library with old books behind her, warm lantern light and violet shadows. Waist-up fashion portrait, no readable book text, no extra people, no logos or watermark.

Generate one image, open the result at full size, and compare it with the attached source. The right-hand tool loads this selected example with its actual reference, prompt and settings. You can try it directly using the tool on the right.

What the actual output shows

The four outputs create clearly distinct costume cues: burgundy velvet, black wings, a violet pointed hat with a brass lantern, and ivory lace with a veil. The short-haired subject remains visually recognizable across the set.

Wardrobe details are interpretations of the prompt, not exact tailoring specifications. The witch result provides a clear face, hat and lantern, so it is the selected reusable example.

Other prompts tested from the same reference

Each result below was generated from the original reference separately, using the same model and settings.

Velvet vampire

Edit image 1 into a tasteful glamorous vampire fashion portrait of the same adult woman. Preserve her recognizable face, skin tone, close-cropped hair and gold studs. Dress her in a fully clothed burgundy velvet high-collar gown, subtle tiny fangs, dark plum lip color. Candlelit stone hall background, rich red and cool blue lighting, confident seductive eye contact without explicit sexuality. Waist-up portrait, no blood, no extra people, text or watermark.

Midnight angel

Edit image 1 into an elegant midnight angel fashion portrait of the same adult woman. Preserve her recognizable face, brown skin tone, short natural hair and earrings. A fully covered structured black satin dress with long sleeves, two large black feather wings behind her shoulders, silver moon-shaped hair accessory. Cool moonlit rooftop background, dramatic but soft rim light, confident editorial pose. Waist-up framing, no nudity, blood, text, logos or extra people.

Ghost bride

Edit image 1 into a graceful ghost bride fashion portrait of the same adult woman. Preserve her recognizable face, brown skin tone, short natural hair and facial proportions. Fully covered ivory lace dress with long sleeves, sheer veil behind rather than covering her face, a small bouquet of pale dried flowers. Abandoned conservatory at dusk with gentle mist and soft silver light. Romantic, mysterious and tasteful, waist-up framing, no wounds, horror gore, extra people, text or watermark.

Check the details before you share

Check skin tone under the new lighting, facial proportions, earrings, fingers and prop grip. Tiny fangs may be subtle, and garment coverage or neckline may differ from a literal description. One run per look is evidence of these examples, not proof of guaranteed four-image consistency.

If a detail needs correction, return to the source or attach the selected result for a clearly scoped revision. State exactly what should change and what must remain. That correction path is workflow guidance; we are not presenting an untested follow-up as a successful result.

Build a four-way audience vote

Use one result per slide at the same crop, then ask viewers to choose a character. Keep the question about styling rather than identity. The witch template in the right panel can be adapted by replacing the reference and changing the wardrobe instructions.

Suggested caption:

One face. Four Halloween identities. Velvet vampire, midnight angel, gothic witch, or ghost bride—which one gets the invitation? AI-generated portraits.

For a different creative task, see Make Funny Halloween Costumes from Everyday Fears with GPT Image 2.5. Start with the loaded example, replace its reference with your own, and adapt one instruction at a time.