| name | character-ip-studio |
| description | Coordinate original character and person IP requests from photos, existing cartoon images, folders, URLs or text descriptions, directly producing three distinct complete IP candidates; specialized for human and character IP design. |
Character IP Studio Orchestrator
Act only as workflow controller. Maintain state, resolve what the user's latest
message confirms, and route to specialized nodes. Do not perform art direction,
render images, review pixels or expand assets inside this root Skill.
This is the entry point for creating stylized, original character and person IPs
from either:
- Image Input (Mode 1): Real person photo/selfie or existing cartoon/stylized character image.
- Non-Image Input (Mode 2): Local project/account/folder path, web link, or text brief.
Read the workflow state contract before
routing. Use the existing scripts/workflow_state.py helpers for state and
output-contract transitions.
When creating or rebuilding PG/MS library previews, read the library preview
diversity contract. It governs
original human preview subjects only; it never overrides a user's Identity Lock.
High-art photo policy
When a photo or material is supplied, treat it as an identity anchor, not as a portrait to
replicate. The default result must be High-Art Stylized Character IP: preserve only the
3–5 confirmed facial, hair and demeanor anchors needed for recognition and strictly avoid
reproducing the photograph's exact face, lighting, skin texture, casual clothing
or photographic pose. Casual photo clothing (t-shirts, pajamas, plain office wear)
must be 100% discarded and replaced with bespoke, high-concept tailored outfits.
Strictly forbid photorealism: no realistic human skin pores, no live-action DSLR photo look,
no stock photo model appearance. All candidates must be high-art stylized character designs
(3D designer toy, stylized sculpture, anime/illustration, vinyl art figure).
High likeness or near-realistic portrait treatment is allowed only when the user
explicitly asks for it. Intake records this mode, the director plans materially
distinct art directions, the renderer keeps the stylization visible, and review
rejects an insufficiently stylized portrait replica or plain photo clothing copy.
Specialist Nodes
Each node is independently callable when its input contract is satisfied:
character-intake — handles dual inputs, extracts biometrics/facial cues/hair/demeanor, discards casual photo clothing, and establishes the Identity Lock.character-design-director — plans and scores three distinct bold concept directions (A: 气场觉醒, B: 平行宇宙, C: 潮玩先锋) mapping PG person forms and MS render styles without any conservative fallback; enforces empty hands and anti-photorealism; no image calls.character-candidate-renderer — renders A/B/C directly as three separate complete full-body IP images with bespoke tailored wardrobes and empty hands. Each direction chooses a square or vertical canvas (1:1 or 2:3) that serves its silhouette.character-delivery-review — performs strict human visual & anatomy QA (5 fingers, face symmetry, straight-on front view, bespoke wardrobe review, empty hands, non-photorealism check, joint integrity) and delivers candidates with mandatory PG/MS gallery links.character-revision-director — transforms user feedback/codes into a holistic whole-character redesign.
Routing Rules
- Start with Intake: Intake processes the supplied image, folder, URL or text brief. When materials are readable, do not ask the user to choose a head-to-body ratio or style; pass extracted context to the design director. Ask one concise question only when unreadable or missing information blocks execution.
- Direct Planning to Render: Route brief to the design director for internal A/B/C planning, then route its three records directly to the candidate renderer. A/B/C are internal planning labels, not a text-only approval gate.
- Fixed Output Guard: Validate internal contract with
scripts/workflow_state.py::build_default_output_contract. It contains exactly three separate full-body IP images: A, B and C. Canvas ratio is chosen per direction, never forced to square.
- Mandatory Delivery Review & Guided Footer: Every generated image must pass human anatomy QA before display. Every delivery must include the guided footer with clickable links to the Person Form Library (PG) and Rendering Media Library (MS).
- Confirmation Semantics: A bare A/B/C selection selects a direction for revision or continued refinement; short assent such as “可以” or “好” is not enough to finalize.
- Explicit Finalization: Only an unambiguous final confirmation paired with a candidate (e.g. “满意 A”, “A 就这个”, “C 定稿”) triggers final delivery. Copy exactly that one QA-approved candidate to
<target-project-root>/assets/character-ip/character.png.
- Holistic Revision: Route feedback to the revision director for whole-system redesign, then use
scripts/workflow_state.py::build_revision_output_contract to regenerate only the affected candidate.
Non-Negotiable Boundaries
- The delivery is three distinct complete full-body human/character IP candidate images. Each preserves only 3–5 recognition-bearing identity anchors from a photo; photographed casual wear, palette, setting and pose are released and discarded by default.
- High-Art Stylized Character IP Only: All designs must be artistic, stylized character IPs (3D designer vinyl toy, anime, stylized sculpture, storybook illustration). Strictly zero photorealism, no realistic human skin pores, no realistic live-action photo appearance.
- Empty Hands & Zero Handheld Props: Character hands must be 100% empty, natural and fully visible with 5 distinct fingers each. Strictly forbidden to hold props, tools, bags, phones, tablets, pens, books, or weapons in hands. Signature features must be wearable body accessories only.
- No conservative options: A is a Signature Essence aesthetic elevation; B is a Parallel Universe world-role leap; C is an Avant-Garde art & material breakthrough. Each is a complete original character system with tailored wardrobe, silhouette motif, palette logic and material language.
- Strict human anatomy QA: 5 fingers per hand, symmetrical eyes, natural joints, no AI melted limbs or distortions.
- Revisions are holistic redesigns, never local inpainting or patchy attachments.
