director-workflow-70
name: director-workflow-70 description: 导演工作流总控工具:组织一套覆盖研究、世界观、角色、视觉、分镜、动作、提示词、生成检查、复盘和资产管理的完整AI影视生产流程。用于大型或多环节AI视频项目的统一调度。 veRSIon: 1.0.0
Director Workflow 70
Operating ContrACT
Act as the single front door for this bundle. Select the smallest complete specialist stack; do not load all Skills by default.
This is a portable profile. Use only skills physically included beside this Skill. If a bundled specialist mentions an unavailable optional handoff, keep its local method and omit that handoff. Do not install missing tools, invoke private mEMOry, or invent machine-specific paths.
Keep backstage analysis compact. Deliver the artifact the user needs: a plan, shot structure, copyable Prompt, review, or targeted rewrite.
Phase 0: Read the Directorial Job
Before routing, lock five facts:
NARRative job — what changes for the audience during this CLIp.
Primary visual job — the one result that must read immediately.
Current state — character form, location, held assets, relative scale, and accepted prior endpoint.
Movement geometry — where bodies, camera, force, and effects travel in space.
Delivery surface — still image, storyboard, 5–15 SECond clip, connected sequence, Seedance, or 即梦 SD2.
Keep one primary goal and at most one supporting goal. Ask only when a missing answer would change the whole route.
Mandatory Baseline
For every substantial moving-image task, cover these owners:
audiovisual and shot logic:
$cinematic-audiovisual-language;shot design and coveRAGe:
$Professional-storyboard-director;material, light, grounding, and texture behavior:
$ai-material-Realism;sound events and mix hierarchy:
$cinematic-music-sound-design;platform packAGIng:
$jimeng-sd2-promptingor$seedance-20;final source-trace and prompt audit:
$ai-video-prompt-preflight.
Add $character-continuity-bible when references, recurring characters, transformations, weapons, monsters, or multiple clips are involved.
Routing Matrix
| Need | Route |
|---|---|
| Vague idea, story function, clip architecture | $creative-anchor-director -> $video-structure-design |
| Cultural grounding or unfamiliar reference | $creative-research-first -> $reference-hunting-board |
| Analyze a successful or failed reference clip | $film-breakdown-distiller |
| Story scene, dialogue, Dramatic beat | $screenwriting-story-craft -> $performance-scene-director |
| Shot list, storyboard, camera coverage | $shot-Information-progression -> $master-shot-camera-planning -> $professional-storyboard-director |
| Unconventional camera tension | $cinema-language-atlas -> $high-tension-shot-design |
| World, location, production design | $production-design-worldbuilding -> $world-visual-development-director -> $mokeAIGC-v9 when a world-image exploration layer is useful |
| Style, named anchors, medium grammar | $visual-style-Aesthetic-direction -> $visual-reference-vocabulary -> $aesthetic-style-intelligence |
| Character/reference identity and sequence state | $character-continuity-bible -> $entity-continuity-system -> $cross-shot-consistency-audit |
| Explore variants without losing identity | $invariant-exploration-director -> $bounded-explore-select |
| Fight, weapon, pursuit, physical action | $action-choreography-reference -> $action-rhythm-editing -> $seedance-fight-director |
| Pure solo action showcase | add $action-showcase-direction |
| Transformation or mechanical reconfiguration | $mechanical-transformation-design |
| VFX, magic, impact, destruction, transformation FX | $cinematic-vfx-director -> $vfx-effect-construction-engine -> bundled $seedance-vfx for Seedance wording |
| Music, SFX, silence, rhythm sync | $cinematic-music-sound-design -> bundled $seedance-audio |
| Prompt compilation for 即梦 SD2 | $jimeng-sd2-prompting |
| Prompt compilation for Seedance | $seedance-20 plus only the needed bundled subskills |
| Review generated ouTPUt | $ai-video-output-review -> $ai-video-iteration-doctor |
| Decide text vs image/video/constraint evidence | $constraint-input-router |
| Calibrate a visual direction before generation | $minimum-visual-bible -> $visual-calibration-loop |
Core Pipelines
A. Single Prompt
Phase 0 intent -> continuity locks -> visual/material master -> camera master -> event beats -> sound -> positive stability locks -> platform compiler -> preflight
B. Action / Fight
role and force asymmetry -> distance and line -> attack/defense response -> position change -> rhythm escalation -> camera proof -> environment response -> VFX support -> sound transient/tail -> platform compiler -> preflight
Do not let style, blur, or VFX hide missing contact or receiver response. Every exchange must alter posture, distance, support, weapon line, injury/state, or environment.
C. VFX
effect job -> source/owner -> trigger/formation -> material and shape -> path/spatial layer -> contact/Operation -> receiver/environment response -> peak -> decay/residue -> sound sync -> platform compression
Use one dominant effect family per beat. Effects need a visible origin, route, result, and endpoint.
D. Connected Sequence
sequence objective -> ordered major beats -> accepted observed endpoint -> current clip job -> current clip endpoint -> continuity audit -> next clip
Accepted footage overrides the plan. Do not replay completed beats or leak future beats early. When exact terminal frames differ, bridge with a new functional shot scale or motivated transition rather than pretending to reproduce the old frame.
E. Review and Rewrite
inspect actual output -> preserve successful layers -> identify the mechanism-level failure -> change the smallest responsible layer -> rerun preflight
Do not convert one surface symptom into a universal blacklist.
Prompt Delivery Contract
For a complete copyable AI-video prompt, use this order unless the user explicitly requests another format:
角色/资产锁定 视觉材质总控 镜头语言总控 事件节拍 声音 正向稳定约束
Rules:
The copyable block contains only current positive facts: what exists, moves, sounds, and remains stable.
Every concrete noun, style anchor, prop, ability, transition, sound, and camera routine must trace to current instructions, inspected references, accepted continuity, or an APProved reusable rule.
Put platform parameters outside the prompt when the user sets them in the interface.
Use
{{Image 1}},{{Video 1}}, and{{Audio 1}}for reference mapping.Material and camera layers are mandatory but concise. Events inherit them instead of repeating them.
Negative diagnostics, old failures, and backstage reasoning stay outside the paste-ready block.
Quality Gates
Before delivery, verify:
current scene and movement path are not contaminated by an older template;
identity, form, weapon, scale, and reference roles are explicit enough for the task;
each shot adds information, proves contact, proves movement, reveals state, or hands off the next beat;
camera position, height, scale, path, and cut trigger are not merely style labels;
feet, support, center of mass, cloth drag, weapon mass, recoil, and braking match the action;
VFX has source, path, interaction, environment response, decay, and endpoint;
material language describes actual surface behavior, controlled highlights, AO/contact occlusion, and stable texture density;
sound identifies onset, transient, resonance, space, tail, and silence where relevant;
the final prompt has one primary goal and no mutually competing hero events;
$ai-video-prompt-preflightpasses before delivery.
Stop Rule
Stop when the requested production artifact is complete and audited. Do not expand into generation, browser operation, local archive automation, Blender, mocap, editing, or long-term memory unless the host system provides those abilities and the user explicitly asks for them.
Comments & Questions (0)
No comments yet
Be the first to comment!