Red Carpet Paparazzi · skill pubblica Open Skill
Red Carpet Paparazzi
Red Carpet Paparazzi
Manhua
2 1 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-10-13/2hfubbsx.jpg
# Red Carpet Paparazzi Turn one uploaded person into that same person caught on a red carpet: couture clothing, event-accurate scenery, and a beautiful stolen-frame photograph (editorial paparazzi, not a posed studio portrait). There is no structured project state. Progress lives in the conversation, canvas assets, and `write_todos`. Image-to-image **must** use only these channels: `i2i.hd.lite`, `i2i.hd`, `i2i.hd.gpt`. Never write a physical model name. ## When to use - User uploads a portrait / full-body photo and wants a red-carpet look. - User asks for paparazzi candids, premiere arrivals, awards-season fashion photos, or high-atmosphere celebrity snapshots. - User wants to change event, outfit, camera style, or regenerate one frame of an existing red-carpet set. Do **not** use this skill for music videos, multi-shot films, or video. This skill produces still photographs only. ## Production stages (references/) Do **not** read every file at once. Call `read_skill_file` for the current stage only. | Stage | File | Tools | When to read | |---|---|---|---| | 1. Intake | `references/01_intake.md` | `recognize_image` | First turn, or a new reference photo arrives | | 2. Live research | `references/02_live_research.md` | `search_web` | After intake, before designing clothes/scene | | 3. Wardrobe bible | `references/03_red_carpet_wardrobe.md` | — | When packing the person into red-carpet clothing | | 4. Venue bible | `references/04_red_carpet_venues.md` | — | When building the carpet, steps, lighting, and crowd | | 5. Candid camera bible | `references/05_candid_paparazzi.md` | — | When writing the photography / stolen-frame prompt | | 6. Generate | `references/06_generate.md` | `generate_image` | When calling I2I | | 7. Edits | `references/07_edits.md` | `generate_image`, `recognize_image` | User wants to change one shot, outfit, or event | Default one-click path: **1 → 2 → 3 → 4 → 5 → 6**. For a single-shot rewrite, read **7** only (plus 3/4/5 if the change needs new visual language). ## Collaboration - **One-click / “just make it”:** run the stages in order. Do not stop between stages to ask for confirmation unless a photo is missing, the subject looks under 18, or the request is sexually explicit. - **Missing photo:** stop and ask for a clear person photo (face visible; full-body preferred). Do not invent a stranger and pretend it is the user. - **Unspecified event:** default to a contemporary **film-premiere / awards-season** mix, then let live research pick the most photogenic current carpet (Oscars, Cannes, Met Gala, Golden Globes, or BAFTA). - **Unspecified framing:** default **portrait** (`orientation='portrait'`). Use landscape only if the user asks, or if the chosen moment is a wide Getty arrival. - Enter the multi-stage path with `write_todos`. At a stopping point or when the set is done, call `set_guidance` in the same turn as the final text. - Failures: report honestly. Never fabricate media URLs. ## Identity lock (non-negotiable) The uploaded person must remain the **same person**. Keep face, age range, skin tone, bone structure, body type, hair color, and distinctive features. Upgrade grooming and clothes; do **not** swap them for a celebrity lookalike. If `recognize_image` suggests a minor, refuse and stop. ## Channel map (hard-coded) | Channel | Role | |---|---| | `i2i.hd.lite` | Fast look test / retry / hidden draft (`add_to_canvas=false`) | | `i2i.hd` | Hero photoreal I2I — fabric, flash, skin, night photography | | `i2i.hd.gpt` | Complex instruction-following I2I — identity + pose + scene in one prompt | Default finished set: **4 photographs**, mixed `i2i.hd` + `i2i.hd.gpt`. Details in `references/06_generate.md`. ## Canvas - Final photos: `add_to_canvas=true`, `entity_type='storyboard_images'`, short `name` such as `Red Carpet Walk`, `Flash Candid`, `Stair Ascent`, `Car Arrival`. - Lite drafts: `add_to_canvas=false`. - Reply in natural language. Do not paste long URLs. ## Notes - Image pro