couple-photo-wardrobe-board

Any number of people outfit change 6-grid

Use this Skill when the user says "outfit change", "change clothes", "restyle", "not satisfied with the costumes", "another version", etc., and the current character photography project already has a clear theme.

Couple is only a two-person preset; the standard data structure of this Skill is participants[], and the number of people can be 1 or any number of multiple people.

Detail Floor

Each candidate set must write the complete outfit for each person one by one, and must not be compressed:

• P01 one complete outfit
• P02 one complete outfit
• ...until everyone

It is forbidden to use "the others are similar", "the remaining members wear the same style", "everyone uniformly wears..." as character costume descriptions. It is allowed to first state the shared motif, but after that the final clothing must still be given person by person.

Process

1. Inherit the confirmed theme, location context, character roster, character relationships, identity references, and avoid items. If the theme is not confirmed, return to kefu。
2. Design 6 group costume combinations that clearly differ in at least two aspects.
3. Each set of JSON must include:
   • name
   • participants: each person includes participant_id、label、outfit
   • group_relationship
   • scene
   • props
4. The six sets must include exactly the same character IDs in the same order; no set may omit or add people.
5. Display the complete text board and wait for the user to confirm image generation.
6. After the user explicitly authorizes it, compile the 6 complete person-by-person costume sets into one generated unnumbered adaptive 6-grid. Each cell must show the complete character roster of that candidate.
   • When generating the 6-grid, do not specify the single-cell aspect ratio, single-cell size, or overall board ratio.
   • The 6-grid does not prescribe the number of columns/rows; the AI may arrange it freely, as long as there are exactly 6 cells, each cell is clear and independent, and the whole is readable.
7. register-grid Only overlay accurate numeric numbering; the model must not generate text or numbers.
8. When the user says "another version", create 6 completely new combinations and continue the numbering.
9. The user may select one or more numbers;select upgrade the selected plan to a confirmed Look, and when a structured location is available, write it into schema v3's participant_looks、visual_lock.participant_outfits and location_lock, then return to paishe。

New JSON example

[
  {
    "name": "LOOK A",
    "participants": [
      {"participant_id": "P01", "label": "姐姐", "outfit": "完整服装描述……"},
      {"participant_id": "P02", "label": "弟弟", "outfit": "完整服装描述……"},
      {"participant_id": "P03", "label": "母亲", "outfit": "完整服装描述……"}
    ],
    "group_relationship": "三人共享米白与灰蓝，但材质与廓形按年龄和角色区分……",
    "scene": "……",
    "props": "……"
  }
]

Old female / male / couple two-person JSON can still be read and automatically converted to P01/P02; do not write the old format for new projects.

Status and boundaries

• WARDROBE_TEXT_REVIEW: text candidate review
• WARDROBE_GRID_REVIEW: numbered grid review
• The new board archives the affected old handoff / shooting plan / Shot boards, and does not delete historical images
• Numbers increase monotonically
• After the theme changes, candidates from the old theme cannot be directly selected into the new process

Commands

python skills/huanzhuang/scripts/wardrobe_board.py create <项目目录> --theme "<confirmed theme>" --options-file <draft.json> --theme-id THEME-001
python skills/huanzhuang/scripts/wardrobe_board.py prepare-grid <项目目录> BOARD-001 --theme-id THEME-001
python skills/huanzhuang/scripts/wardrobe_board.py register-grid <项目目录> BOARD-001 --image <六宫格.png> --theme-id THEME-001
python skills/huanzhuang/scripts/wardrobe_board.py select <项目目录> --ids 1,3 --theme-id THEME-001

Cross-Session / theme restart / outfit change without quality degradation

Outfit change is a costume branch, not a "simplified mode". Whether it is a new session, a restarted Theme, or "another version", each candidate must still fully write out all people, and must not compress the later people into "the others are similar".

The scene must maintain the confirmedspatiotemporal anchor: in the candidates, scene must not abbreviate a real location or an in-work location as "hotel / street / courtyard / city". At the outfit change stage, only the location anchor needs to be retained, and area-level candidates may be given; do not lock in Shot-level window positions, corridor corners, steps, or shoreline nodes in advance.

Real indoor locations

Real indoor outfit change plans must not write only the city or neighborhood. Prefer to retain a real named hotel/building; private apartment concepts must retain at least a real street/route anchor, and be marked as conceptual_private_interiorIt must not fabricate door numbers or room numbers.

2.24.20 Hairstyle Boundaries

• The Wardrobe Board is only responsible for clothing/overall styling preview; it does not lock the official shoot hairstyle.
• If the user truly selects a new outfit change plan, that "change outfit" action invalidates the original official hairstyle/makeup according to the Remix Dependency Graph; the system should immediately rerun the Editorial Hairstylist → Editorial Makeup Artist, and then lock the new Look. The temporary makeup and hair of the outfit preview itself does not directly become the official result.

2.24.20 Makeup Boundaries

• Makeup in the Wardrobe Board is only used for visual completeness/getting a feel; it does not lock the official shoot makeup look.
• Once a new outfit change is confirmed, immediately recalculate the official makeup and hair; after that, downstream only inherits the new lock.