Soft Studio Portrait · skill Open Skill công khai
Soft Studio Portrait
Soft Studio Portrait là một Open Skill công khai trên Digen. Mở ra để xem workflow này tạo gì.
Other
1 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/9sj8zvgu.jpg
# Soft Studio Portrait (IMG2.5)
Produce one photorealistic, text-free, front-facing studio portrait of a single adult woman from a short text brief, optionally guided by one reference image that supplies her identity. Output is a single image with a film-still look on a dark gradient studio background, framed 3:4 vertical.
## When to use
- User asks for a portrait, headshot, profile photo, "studio photo", "cinematic portrait", "film-still portrait", or a person image in this signature style.
- User supplies one reference image of a person and wants the same person rendered in new studio styling (different outfit / hair / makeup / earrings / theme).
- User wants to iterate on a portrait already produced in this conversation (brighter, warmer, different wardrobe, different expression).
Not for: full scenes, groups, storyboards, product shots, or video.
## Input handling
Pick exactly one generation mode before writing the prompt:
| Input | Mode | Tool call |
|---|---|---|
| Text brief only | Text-to-image | `generate_image(model='t2i', ...)`, no `input_urls` |
| Text brief + one reference image | Reference-guided image-to-image | `generate_image(model='i2i', input_urls=[reference image], ...)` |
- Reference images may be given as an uploaded asset, an `asset_id`, or an `s3://` / https URL from this conversation. Never invent a URL.
- If several images are attached, use the first one that clearly shows a single person's face as the identity reference.
- If the user provides a reference image but explicitly asks to keep the clothing, background or lighting of that image, tell them this skill keeps only the identity and restyles everything else.
## Steps
1. Determine mode (text-only → `t2i`; reference present → `i2i`).
2. Write the image prompt in **English**. Obey the prompt contract below: every mandatory element must be present, in this order. Keep it one dense paragraph, roughly 60–120 words; do not paste the user's raw text and do not output JSON.
3. Call `generate_image` with the parameters in "Parameter defaults".
4. Reply in the user's language with one or two sentences describing the result (styling choices, what can be changed next). Do not paste URLs or the raw prompt back.
5. If the user then asks for a variation, keep mode, background, light direction and framing identical, and change only the styling elements they named.
## Style prompt reference
The house style is pinned as a copy-ready prompt in `references/style-prompt.md`: the fully written English master prompt with `<placeholders>`, the identity-preserving variant for `i2i`, filled-in example prompts, the theme-variation table and the verbatim negative clause.
Read it with `read_skill_file('soft-studio-portrait', 'references/style-prompt.md')` before writing the prompt in step 2, and again whenever the user asks for a restyle or an iteration. The blocks below are the contract that prompt must satisfy — the reference file supplies the exact wording.
## Prompt contract (mandatory elements)
Build the prompt from these blocks, in order:
**1. Framing / composition** — "text-free front-facing portrait of one adult woman", "centered head-and-shoulders composition", "vertical 3:4 frame", "camera at eye level, direct gaze", "shallow depth of field, 85mm portrait lens look".
**2. Light** — "soft cinematic studio light with a single directional key light from the side (about 45 degrees), gentle fill and soft falloff", "soft shadow on the far side of the face", "no harsh specular highlights". Light must always read as *directional*, never flat or ambient-only.
**3. Background** — "seamless dark gradient studio background, near-black at the edges lifting to deep grey behind the subject". No props, no furniture, no visible equipment, no second light source in frame.
**4. Skin / rendering** — "finely detailed natural skin with visible pores, soft shadow transitions, subtle color grading, film-still look, unified muted palette, consistent white balance".
**5.