ııı motion.mdVideo style specification
ElevenLabs / V300

Flows Agent — describe it, the agent wires the workflow

Let natural language grow into an editable media workflow. Nodes represent real outputs and lines represent dependencies; later edits replace only related parts.

Composition and choreography Playable · Replaceable content
Overall choreographyRule example
0.00 / 48.67 seconds
Motion and composition example Open separately ↗
View HTML source
<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Flows Agent — describe it, the agent wires the workflow · sequence</title><style>html,body{margin:0;width:100%;height:100%;overflow:hidden;background:#f7f6f0}canvas{display:block;width:100%;height:100%;object-fit:contain}</style><body data-mode="sequence"><canvas width="1000" height="600"></canvas><script src="data.js"></script><script src="common.js"></script><script src="scene.js"></script><script src="bridge.js"></script></body></html>
Visual elementsInput barTask listImage nodeVideo nodeAudio nodeDependency connectionVariant groupFinished output
MOTION LANGUAGE

Motion and composition

Local motion · Composition rules

01

Turn input into tasks before nodes

After a brief prompt box, list image, video, voiceover, or editing tasks, then place them on the canvas in dependency order. Keep node titles and states clear; do not show unfinished results as final outputs.

Keep pale nodes and fine lines readable, leaving orange fluid or soft glow outside the workspace. Use clear sans serif headings, with smaller media types, states, and parameters.

02

Use connections to show asset flow

Images may feed motion nodes, voiceover enters editing, and editing produces the final video. Activate downstream nodes only after completion; arrange complex graphs along content flow and minimize crossing lines through text.

Node relationshipProduction constraints
Reference→imagePreserve subject identity
Image→videoBuild motion around the existing subject
Sound→editingMatch duration to visual sections
Editing→finished videoWait for required media to be ready
03

Preserve other work during local edits

Preserve other work during local editsRule example
0.00 / 15.00 seconds
20–35 seconds · Motion example Open separately ↗
View HTML source
<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Flows Agent — describe it, the agent wires the workflow · sequence</title><style>html,body{margin:0;width:100%;height:100%;overflow:hidden;background:#f7f6f0}canvas{display:block;width:100%;height:100%;object-fit:contain}</style><body data-mode="clip" data-start="20" data-end="35"><canvas width="1000" height="600"></canvas><script src="data.js"></script><script src="common.js"></script><script src="scene.js"></script><script src="bridge.js"></script></body></html>

When replacing a product, changing aspect ratio, or selecting voiceover, highlight relevant nodes and retain unaffected nodes and links. Then show matching variants so viewers can see the scope of the change.

04

Show the work and the workflow at the end

Show a complete ad segment first, then pull back to the node overview. Replace product, prompt, media, and sound together for a new theme; preserve the structure from planning and dependencies to local edits and final output.

SEQUENCE

Sections and rhythm

Click a section to seek to it in the choreography example above.

Composition reference at 0.00 seconds
0.00s
Composition reference at 9.00 seconds
9.00s
Composition reference at 17.00 seconds
17.00s
Composition reference at 28.00 seconds
28.00s
Composition reference at 35.00 seconds
35.00s
Composition reference at 44.00 seconds
44.00s

Visual references · Discrete still frames

KEEP THE RELATIONSHIPS

Creating other themes

  • Replace the advertised product while preserving plan→dependencies→local edits→finished output
  • Node types must match real media dependencies; connections must not be merely decorative

Production constraints

Input bar / task list / image node / video node / audio node / dependency connection / variant group / finished output

  • Convert natural language into node dependencies first, and make output nodes wait for upstream assets to finish
  • subsequent edits replace local nodes while preserving other connections to demonstrate editability
  • choose aspect ratio and audio before the finished output, then pull back to show the full workflow