Halloween Portrait Makeover · সর্বজনীন Open Skill
Halloween Portrait Makeover
Halloween Portrait Makeover
Auto
1 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/ch6zx4bd.jpg
# Halloween Makeover (reference photo → Halloween portrait)
Turn the person's photo the user uploaded into a single Halloween-costume portrait quickly and cheaply, using reference-based image generation: **the face and identity stay exactly the person in the uploaded photo — only the costume, pose, and scene change**. The output is one image; there is no structured project state.
## When to use
- The user uploads or pastes a photo of a person and says "turn me into a Halloween look", "Halloween costume change", "Halloween makeup", "make a Halloween portrait out of this photo", "make it a Halloween avatar".
- The user names a specific look: witch, vampire, pumpkin spirit, ghost, zombie, catwoman, Frankenstein, skull makeup, etc. — this only affects the "Costume" block of the prompt below.
- The user only wants to change one element ("swap the background to an old porch", "change the hat to a witch hat", "make the light warmer") — replace only that block and copy the rest of the previous prompt verbatim; do not redo the whole image.
## Input and output
- **Input**: one photo of a person (an image asset uploaded in this turn, or an existing image URL on the canvas). 1 image is best; if the user gives 2, use the first as the primary identity reference.
- **Output**: 1 image, `entity_type='halloween_portraits'`, portrait orientation by default.
- If no reference photo is given: ask for one first, clearly stated in a single request (front-facing, sharp, face unobstructed). Do **not** invent a "Halloween person" out of nothing.
## Steps
1. **Confirm the reference photo**: get the image reference for the user's photo (use the canvas asset directly if it exists). If there is no image, ask for one first and state the requirements in one go (front-facing, sharp, single person).
2. **Confirm the look**: if the user did not specify, default to "witch" and offer 2–3 alternatives in the reply (vampire / pumpkin spirit / ghost). Do **not** interrogate the user about makeup, props, background, and aspect ratio item by item; one round of questions is enough.
3. **Assemble the English prompt**: strictly concatenate the four blocks from "Prompt conventions" below. If the user is happy with the previous version and only wants one change, replace only that block.
4. **Call `generate_image`** with:
- `prompt` = the English prompt assembled in step 3
- `input_urls=[reference image]` (the reference image goes only here; do not paste it into the prompt text)
- `model='i2i'` (default: has a reference image, fast, low cost); switch to `model='i2i.hd'` only when the user explicitly asks for higher quality (`.hd*` does not support NSFW, which a Halloween costume shot does not need)
- `orientation='portrait'` (default); pass `'landscape'` when the uploaded photo is horizontal or the user explicitly wants a horizontal version
- leave `resolution` unset (platform default 720P); pass `'1080P'` only when the user explicitly says "high-res / for print"
- `name` = a short title (e.g. `halloween-witch`); `entity_type='halloween_portraits'`
- do not pass pixel width/height; do not pass `batch_size`
5. **Reply**: one or two natural-language sentences stating what changed (costume + scene + lighting). **Do not paste the image URL.** In the same turn you may call `set_guidance` with 2–3 next steps based on the current output ("switch to the vampire look", "change the background to a foggy forest", "try a brighter makeup look"), then end the turn.
> Low-cost principle: one image per version by default. Call the tool again only when the user asks for "another version / a different one"; do not silently run several images to pick from.
## Prompt conventions (fixed)
**The prompt is always in English**; any character names or spoken words (e.g. "vampire") keep their English wording. On every call, concatenate the four blocks below into **one single English prompt**, in this order.
### Block 1: Subject and identity preservation (always requir