This skill should be used when the user asks for body action, choreography, physics, object movement, movement timing, action continuity, stunt direction, or mo
复制下面这句话,粘贴给 Claude Code、Codex、Cursor 等 AI 编程工具,它会读取安装说明并在你确认后完成安装。
请阅读 https://ai.atlankj.com/install/asset/gh-seedance-motion-e0709f2078c3 ,按照其中的说明把「seedance-motion」安装到你(当前 AI 工具)中。执行前先告诉我将运行的命令和写入的位置,等我确认。
查看 AI 将读取的安装说明正在读取 GitHub 原文…
内容来自 GitHub 原始文件,由原作者维护。在 GitHub 查看
Before producing prompt text, a prompt-ready block, a rewrite, an example, or a compiled clip, load the Director's Read, classify the brief, and complete its canonical narrative or non-narrative record. Translate that record into visible or audible carriers and keep its internal labels out of final generation prose.
Use physical verbs and consequences. Motion should be observable on screen, timed within the clip, and assigned to a subject or object. Prefer one strong action with a visible endpoint over several vague actions competing for attention.
Load reference-workflow for video-motion references, shot-list-continuity for action handoffs across shots, examples-by-mode for safe edit, extend, and R2V patterns, and directing-engine when motion is performance: translate the scene's emotion into one true visible gesture per beat - a playable action with an objective and subtext - instead of an emotion word the model cannot render.
Motion is the verb of the user's story - the thing they came to see HAPPEN. The soul here is consequence: motion that begins, lands, and changes something feels lived; motion that loops feels generated. Every action carries the story one beat forward, or it doesn't belong in the clip.
State: actor/object, action, force level, timing, physical consequence, continuity requirement, and endpoint.
| Motion type | Strong phrase | Weak phrase |
|---|---|---|
| Subtle acting | Character A inhales, grips the cup tighter, then sets it down without looking away | she feels nervous |
| Product material | condensation beads gather, merge, and slide down the bottle neck | the product looks refreshing |
| Choreography | Character B ducks under the swinging bag, pivots left, and stops in a guarded stance | fast action fight scene |
| Object physics | paper receipt lifts in the fan breeze, flips once, and lands face-up | papers move dynamically |
| Environmental motion | rain streaks diagonally across the backlight while puddle ripples spread from footsteps | stormy weather atmosphere |
Official material claims strong physics; extract it by writing causes and letting the model compute consequences (field-observed emphasis - test before promising results). State mass, force, and material, then name one consequence the camera can see: the heavy oak door swings shut and the candle flames bend toward it beats the door closes dramatically. Consequences prove the action: weight shows in landing compression, momentum in overshoot and recovery, friction in skid length, wind in what it displaces. One physical cause with two or three visible consequences reads stronger than three separate actions.
Use a three-beat structure for short clips: setup, action, changed end state. Example: 0-2s: candle flame steady; 2-4s: door opens and flame bends; 4-6s: smoke trail curls toward the hallway. Time segmentation is useful for action, VFX, lip-sync, and product demonstrations, but avoid frame-perfect overload unless the user truly needs it.
When sound drives the motion, pair each visible change with one beat or SFX: door click at 2s, light pulse on the downbeat, hand releases the cup on the final chime. Do not ask for many cuts, locations, and micro-actions inside one short clip.
For reference footage, use only owned, licensed, public-domain, stock, mocap, rehearsal, or self-recorded material. Map @Video1 to motion, camera, timing, or blocking, not identity, unless the identity is authorized. If a reference contains a real person, transfer only general motion or camera behavior and explicitly exclude likeness transfer.
Hands, faces, logos, and product geometry drift when too many actions occur. Reduce motion around fragile details: lock the camera for lip-sync, keep hands in simple poses, ask product parts to remain rigid, and move light or environment instead of the core identity anchor.
When sequence state is present, inherit the observed action phase, open motion vector, current clip scope, continuity locks, exact reference tags, and reserved future beats. Do not replay actions marked already happened or completed. Do not perform a reserved beat early; carry unfinished motion from the accepted end state into the next clip.
Return the motion phrase, timing pattern, reference role map if any, and repaired prompt language.