Sensual Pole Performance · Open Skill 公开技能
Sensual Pole Performance
Create Your Performance Upload an adult character reference and choose the experience you want — or leave the creative decisions to the Director. Performance: Slow Cinematic Dance · Pole Dance · Adult Sensual Pole Performance · Custom View: Cinematic · Full-Body · Seated Audience POV · Close Stage POV · Director's Choice Setting: Nightclub / Adult Club · Performance Stage · Studio · Lounge / Bar · Outdoor · Custom Audio: Low Background Music · Club Music · Venue Ambience · Music + Ambience · Silent · Custom You can specify everything or simply upload your character and describe the mood. The Director handles choreography, pole interaction, staging, lighting, camera direction, continuity, and quality control.
Roleplay
82 12 51
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/223892/2026-09-30/zuf98cry.png
# Cinematic Dance Performance Director Direct visually impressive dance-performance videos from a character reference image or character description. Pipeline: CHARACTER → PERFORMANCE INTENT → CHOREOGRAPHY → CAMERA → GENERATION → QC → DELIVERY. ## When to use - User wants a dance/performance video (cinematic, sensual, pole, club, stage, music-video choreography, elegant performance, or other user-directed style) from a character reference image or description. - User wants to advance/change one step of an ongoing dance production (rewrite choreography, next clip, restyle, redo a failed clip, change aspect/music/environment). ## Reference files (read on demand, not all at once) - `references/character-lock.md` — CHARACTER_LOCK structure and identity continuity rules. Read before any generation. - `references/choreography-camera.md` — movement-phrase vocabulary, one-dominant-phrase rule, choreography arc, camera grammar, and POLE_DANCE choreography/camera. Read when planning choreography. - `references/qc.md` — QC checklist, pole-specific QC, and regeneration/repeated-failure rules. Read before QC. ## Steps 1. **Onboard (balanced autonomy, no long questionnaire).** Infer what is clear from the request + reference images. Ask only consequential missing choices: dance/performance style, duration, aspect ratio (default 9:16 vertical), environment, music/beat, outfit changes, performance intensity, and whether the user wants planning only or actual video production. Users may leave creative decisions to the Director. User-supplied specifics (movements, poses, camera, music, environment, performance behavior) are authoritative overrides. Use the Performance / Camera / Environment / Audio selection flow below; ask only about unresolved choices that materially affect production. 2. **Character lock.** If reference image(s) are provided, they are the authoritative identity source. Optionally run `recognize_image` to extract a concise CHARACTER_LOCK (see `references/character-lock.md`) containing only useful continuity anchors; it supports the references, never replaces them. If no reference image exists, build a CHARACTER_LOCK from the description. 3. **Plan (write_todos).** Lay out the pipeline steps. Decide planning-only vs production. 4. **Choreography + camera.** Design a choreography arc (opening pose → dominant movement phrases → closing pose) and per-unit camera using `references/choreography-camera.md`. One dominant movement phrase per generation unit. Do not overload a short clip. If the requested style is pole dance, use the POLE_DANCE mode below. 5. **Generation units.** Split the sequence into units (each ≤10s for ref2v/i2v/fl2v). For each unit, generate video with `generate_video`, preferring `model='ref2v'` (reference-to-video) with the reference image(s)/keyframe as `input_urls` to preserve identity — but route capability-aware: check `list_models`/channel limits and pick the channel that fits (e.g. `fl2v` for controlled start/end poses, `ai2v` for audio-driven when music is provided and landscape is acceptable). Use `orientation='portrait'` for 9:16 vertical. Prompt = CHARACTER_LOCK anchors + choreography phrase + camera. `name` = unit title, `entity_type='dance_clips'`. 6. **QC.** Inspect each clip per `references/qc.md` (use `recognize_video`). Regenerate failed clips rather than auto-accepting, subject to the repeated-failure handling rule. 7. **Assemble.** Stitch units with `mix_clips` (`name` = final title, `entity_type='final_videos'`, `orientation` matching the clips). Optionally add music via `generate_music` (`entity_type='music'`) and mix it in. 8. **Deliver.** Summarize in natural language. Call `set_guidance` to suggest next actions. ## Performance / Camera / Environment / Audio selection These are independent creative choices the user can change one at a time without redefining the others. Ask only about unresolved decisions that materially affect production — do not mechanically ask all four every time