Luxury Night Street Fashion Portrait · Open Skill 공개 스킬

Luxury Night Street Fashion Portrait

Luxury Night Street Fashion Portrait은 Digen의 공개 Open Skill입니다. 열어 보면 이 워크플로가 무엇을 만드는지 알 수 있습니다.

@GG_Bone

Story

0 0 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/tjyvp3mc.jpg

# Luxury Night Street Fashion Portrait

One-image delivery: turn the user's text brief (plus an optional 1–3 reference images) into a **3:4 vertical** cinematic night street fashion portrait. A single generation, no multi-stage flow.

- Input: text (subject, occasion, person and outfit description) + optional reference images (the person / the evening outfit / the night street)
- Output: one photorealistic portrait image, automatically added to the canvas

## When to Use

- The user asks for "a night fashion portrait / a night-street evening-wear shot / a cinematic night photo" and provides a brief or reference images.
- The user uploads a photo of themselves and asks to keep the likeness while changing the outfit or the scene.
- The user uploads only one reference image (person only / clothes only / location only), or provides no reference image and wants a text-only generation.

## Input: Three Kinds of Reference Images, Independent of Each Other

Reference images are **not required**, at most 3, and each one is responsible for exactly one thing — the skill still works if any of them is missing:

| Reference image role | What it does | How to fill it in from the text brief |
|---|---|---|
| Person reference | Facial features, bone structure and identity **stay unchanged** | Add age and vibe, hairstyle, makeup, expression |
| Evening outfit reference | Reproduce the whole look (colour / material / cut / accessories / shoes and bag) | Describe the material and colour of the gown, suit or coat |
| Night location reference | Reproduce the architecture, lighting and perspective of the street / porch / entrance | Describe the concrete place elements (door lantern, stone steps, wet asphalt) |

If you cannot tell which category an image belongs to, use it for the closest matching role, **do not ask the user for more images**, and fill the remaining slots from the text.

## Steps

1. **Read the brief**: extract "who she is → what she wears → where she stands → what pose and mood" from the text. Fill in reasonable gaps when the input is thin (e.g. only "before a gala dinner" → hotel entrance + evening gown + glancing-back stance); do not interrogate the user for details.
2. **Assemble the English prompt**: use the fixed skeleton in "Prompt Conventions" below plus the subject slots, written as natural English prose in one paragraph — not parameter-style fragments.
3. **Call `generate_image`** (channel is hard-coded; never pass a physical model name):
   - With any reference image: `model='i2i.hd'`, `input_urls=[...]` (order: person → outfit → scene, matching the order described in the prompt)
   - Without any reference image: `model='t2i.hd'`, do not pass `input_urls`
   - Fixed parameters: `orientation='portrait'`; when an exact 3:4 vertical is needed also pass `aspect_ratio='3:4'`; `name=` a short title from the brief (e.g. "Rainy Porch Velvet Gown"); `entity_type='portraits'`
   - `resolution` is not passed by default (720P); pass `'2K'` when the user asks for high definition — anything above the model's limit is automatically downscaled by the platform
4. **Reply**: one or two natural-language sentences about what the shot shows, **do not paste URLs**. When a next step is worth suggesting, call `set_guidance` in the **same turn** with 2–3 concrete options (change the evening outfit / change the night location / same scene, different pose).

## Prompt Conventions

**Fixed part** (copy verbatim every time, so that every image from this skill shares the same look):

```
LIGHT: warm directional key light on the face and outfit, cool night ambient highlights rimming the hair and shoulders, deep cinematic shadow falloff, unified amber-and-teal colour grade, background street lights rendered as soft bokeh orbs.
CAMERA: 85mm lens, f/1.8, eye-level, shallow depth of field, medium to three-quarter length framing, 3:4 vertical, subject slightly off-centre, headroom kept tight.
STYLE: film still, ultra-realistic photography, natural skin textu