Ten original looks · image + video prompts

Halloween Lookbook 2026

Quick answer

This lookbook turns ten Halloween costume ideas into copy-ready image and video prompts. Pick a look, upload one clear photo to an editing model, keep the face and body consistent, and change only the costume, lighting and set. Every guide names a low-cost route through Kie, WaveSpeed, Fal, Replicate or SeeAPI.

Moonlit garden path with carved pumpkins, violet foliage and warm lantern light
Original AI illustration for the 2026 Halloween Lookbook.

Halloween searches arrive late and move fast. A costume idea is useful only when it can become a picture or a short clip before the trend passes. That is the job of this lookbook: ten original looks, each with a still-image prompt, a five-second video prompt and a model route that does not require a studio.

The guides are written for real photos. They tell the model what to preserve first, then what to change. That order matters: identity, body proportions and the number of people in the frame are the parts viewers notice immediately when they go wrong.

Looks10 original costume guides
PromptsImage, video and negative prompts
RoutesKie, WaveSpeed, Fal, Replicate, SeeAPI
FormatStill portrait or 5-second clip

The ten looks

Each card opens a single-language guide with its own prompt recipe, editing checklist and FAQ. The set is deliberately varied: solo portraits, a couple, two pet looks and one location-led video.

Fictional adult with long violet braids, silver hair accessories and a black velvet gothic costume in a moonlit garden Midnight Braids Midnight Braids is a gothic Halloween look built from long violet braids, silver hair jewellery and a black velvet outfit. The safest AI edit keeps the face, skin tone and body shape, changes only the hair, clothing and moonlit garden, and uses a vertical crop for social video. Read the guide Fictional adult in a midnight-blue celestial gown with silver star embroidery and a moon crown inside a moonlit observatory Cosmic Princess Cosmic Princess is a celestial Halloween look: a midnight-blue gown, silver star embroidery and a delicate moon crown. The best AI edit keeps the subject's face and pose, adds glow only to the fabric and background, and avoids turning the skin into plastic. Read the guide Original plush haunted unicorn mascot costume with a spiral horn and amber eyes in a foggy room Haunted Unicorn The haunted unicorn is an original mascot-horror look: a plush white hood, a spiral horn and small amber lenses, photographed in fog. Keep the design generic, show the human face where the costume allows it, and avoid copying any film or game character. Read the guide Fictional adult in a wearable robot costume with metallic panels and blue LED accents, human face visible Wearable Robot A wearable robot costume uses metallic panels, articulated shoulder pieces and small blue LEDs, while the human face stays visible. The goal is a believable costume, not a new mechanical character; the prompt must say that the face is not replaced. Read the guide Small dog in a soft pumpkin costume with a green leaf collar in a warm autumn studio Pumpkin Pet The Pumpkin Pet look dresses a cat or dog in a soft pumpkin costume with a small green leaf collar. The edit must preserve breed, fur colour, markings, eye colour and paw anatomy; the costume should sit naturally and never cover the eyes. Read the guide Fictional adult with split-face half-skull Halloween makeup, one human side and one painted bone side Half-Skull Skeleton Half-skull makeup splits the face: one side stays human, the other becomes painted bone. The AI edit must preserve the human half exactly, keep the eyes aligned, and avoid turning the whole face into a skull mask that no longer looks like makeup. Read the guide Fictional adult in a moss-green forest witch costume with an antler crown and a lantern in a foggy woodland Forest Witch The Forest Witch is an earthy Halloween look: a moss-green layered dress, an antler crown, leather boots and a small lantern. The AI edit should preserve the face and body, keep the silhouette readable, and use fog and one warm light instead of a busy forest. Read the guide Two fictional adults in coordinated cowboy and angel Halloween costumes at warm desert dusk Cowboy & Angel Cowboy and Angel is a two-person costume: one warm, western silhouette and one soft, feathered look, tied together by a shared cream-and-sand palette. The AI edit must use two reference photos, preserve both faces, and keep exactly two people in the frame. Read the guide Empty yellow-lit corridor with damp carpet, low fog and a distant silhouette Backrooms Fog The Backrooms look is a liminal-space video: an empty yellow-lit corridor, damp carpet, low fog and a distant silhouette. It needs no costume. Keep the location generic, avoid named entities, and animate the camera and fog rather than a character. Read the guide Small dog in a retro sequin disco costume beside a warm mirror ball and amber lights Disco Pup Disco Pup is a retro pet look: a small sequin vest, a stretch headband and warm mirror-ball light. The edit must keep the pet's breed, fur pattern and eye colour, and the costume must sit comfortably without covering the ears or eyes. Read the guide

How to use this lookbook

  1. Choose a look that matches the photo you already have, not the one that looks best in a thumbnail.
  2. Run the image prompt on a low-cost model and compare two outputs before paying for video.
  3. Check the face, hands, hairline and the number of people. Fix those before adding more scene detail.
  4. Animate one small movement. A head turn, a blink or a fabric move is enough for five seconds.
  5. If the result is going on a profile grid, keep the same palette across the set so it reads as one project.

Model routes at a glance

The same prompt behaves differently across providers. Use this table as a starting point, not a guarantee: model names, aliases and prices change, and a provider may route to a different backend on any given day.

ProviderImage modelVideo modelBest at
Kie.aiGPT Image 2 (image edit)Kling 2.6Face preservation, multi-reference edits, quick iteration
WaveSpeedSeedream 5.0 Flash Edit / Pro EditWan 2.6 Image-to-Video FlashLow-cost batch edits, fabric and scene replacement
FalFLUX Kontext ProKling / Wan endpointsTargeted local edits and clean compositing
Replicateopenai/gpt-image-2 or FLUX KontextKling / Wan community modelsOne API for many models and easy comparison
SeeAPIOpenAI-compatible image routeKling / Wan routeFallback capacity when the primary route is busy

Prompt studio

The studio assembles a prompt from the look, the subject and the output type. It does not call a paid model from this archive; it gives you a clean prompt and a route so the generation happens in the tool you already pay for.

From the BFE Halloween archive

These looks are a seasonal companion to the archive. The Banned from Equestria Halloween Special is a separate playable build with Pinkamena and Big Mac content, and the final Alpha 1.5 build is the version most guides describe.

FAQ

Are these prompts free to use?

Yes. Copy and adapt them for your own photos. Model usage, output rights and commercial use are governed by the provider you choose.

Can I use a phone selfie?

Yes, if the face is sharp and evenly lit. A front-facing photo with the hairline and shoulders visible is more useful than a high-resolution photo with a hand across the face.

Why do the guides name five providers?

Because availability and price change. Naming several routes lets you switch when a model is busy, deprecated or too expensive for a test.

Do the preview images show real results?

No. They are original AI illustrations made for the guide. A prompt is a starting recipe; your photo, model version and settings determine the result.

Is this affiliated with Banned from Equestria?

No. The lookbook is an editorial companion on the same unofficial archive. It is not endorsed by the game's creator or any model provider.

Image and prompt disclosure

Preview images on this page are original AI illustrations made for the guide. They are not game screenshots and not results from a reader's upload. Prompts are starting recipes; model behaviour, price and output quality vary by provider and date.