Skip to content
FunCoding

Search

Search docs, Skills and MCP

ltx-director

Drive the LTX Director (Timeline) node. Its Add Image/Text/Audio buttons are DOM-only and cannot be clicked by an agent; edit the hidden timeline_data JSON widget instead. Load when a workflow contains LTXDirector / LTXDirectorGuide / PromptRelayEncodeTimeline, or when asked to add, move, retime, or remove timeline segments (image / text / audio / motion).

AI 与智能体797plugin/skills/ltx-director/SKILL.md

Install

Send this to Claude Code, Codex or Cursor. The agent checks the Skill for safety first and installs it only after you confirm.

读取 https://funcoding.ai/skills/artokun/comfyui-mcp/ltx-director/install.md ,按里面的步骤帮我安装这个 Skill。

SKILL.md

LTX Director (Timeline)

LTXDirector (pack: WhatDreamsCost-ComfyUI, category WhatDreamsCost) is a video timeline editor node. Its on-canvas UI has Add Image, Add Text and Add Audio buttons.

You cannot click those buttons. Don't try.

They are not node inputs. The pack builds them as raw DOM elements with JS handlers:

addTextBtn.addEventListener("click", () => this.addTextSegmentFreeSpace());

Panel/MCP tools drive the LiteGraph node model (widgets + inputs). They cannot invoke arbitrary DOM handlers, so "click Add Text" is impossible. This is a hard limitation, not flakiness. Say so plainly rather than retrying.

The real control surface: timeline_data

Everything those buttons do is serialized into ONE string widget, timeline_data. The pack's own code treats it as "the absolute source of truth". Set it and the editor renders it.

It is listed in the pack's HIDDEN_WIDGET_NAMES, so it does not appear in the node's visible widget list, but it is an ordinary input and is settable:

create_workflow(action="modify", workflow=<the graph>, operations=[
  { op: "set_input", node_id: "42",
    input_name: "timeline_data", value: "<json string>" },
])

The value is a JSON string, not an object. Empty state is "{}".

Sibling hidden widgets: local_prompts, segment_lengths, guide_strength, audio_data, use_custom_audio, inpaint_audio, use_custom_motion, override_audio.

Schema (verified against a real production workflow)

{
  // TRACK GATES — a track's segments are IGNORED unless its gate is true.
  "mainTrackEnabled": true,
  "audioTrackEnabled": false,
  "motionTrackEnabled": false,

  // must mirror the node's start_frame / duration_frames widgets (see Edges)
  "normalStartFrame": 0,
  "normalDurationFrames": 4393,

  "global_prompt": "",
  "overrideAudio": false,
  "inpaint_audio": false,

  // retake mode
  "retakeMode": false, "retakeStart": 24, "retakeLength": 48,
  "retakePrompt": "", "retakeStrength": 1, "retakeVideo": null,
  "retake_global_prompt": "",

  // display-only
  "propHeight": 90, "globalPropHeight": 60, "showFilenames": true,

  "segments": [],        // main track: image AND text segments
  "motionSegments": [],
  "audioSegments": []
}

Text segment (segments)

{ "id": "seg1", "start": 0, "length": 48, "prompt": "wide shot, neon city", "type": "text" }

Image segment (segments)

{
  "id": "1784657553220x4fxu",
  "start": 0,
  "length": 241.68,
  "prompt": "",
  "type": "image",
  "imageFile": "whatdreamscost/Nintendo_ZeldaUrbosaHotspringKF_00002_.png",
  "imageB64": "/api/view?filename=Nintendo_ZeldaUrbosaHotspringKF_00002_.png&type=input&subfolder=whatdreamscost"
}

imageB64 is a misnomer. It holds a /api/view URL, not base64. So an image segment points at a file already in ComfyUI's input dir. Full agent-drivable recipe:

  1. upload_image (action:"image") → puts the file in the input dir (note its subfolder/name)
  2. imageFile = "<subfolder>/<name>.png"
  3. imageB64 = "/api/view?filename=<name>.png&type=input&subfolder=<subfolder>"

Audio segment (audioSegments)

{
  "id": "17846562234967gjvk", "type": "audio",
  "start": 0, "length": 4393, "trimStart": 0, "audioDurationFrames": 4393,
  "audioFile": "whatdreamscost/Intergalactic Hip Hop.mp3",
  "fileName": "Intergalactic Hip Hop.mp3",
  "waveformPeaks": [0.047, 0.274, 0.537, "…~200 floats…"]
}

waveformPeaks is the rendered waveform. It is cosmetic. Omit it or pass [] if you're writing a segment programmatically; the audio still plays. Don't fabricate plausible-looking peaks and imply they were measured.

Edges (get these wrong and it silently does nothing)

  1. Track gates. Pushing into audioSegments does nothing while audioTrackEnabled is false. Set the gate in the same edit. (The reference workflow ships an audio segment with the track OFF, which is easy to misread as broken.)
  2. Frame bookkeeping must agree. normalDurationFrames == duration_frames widget == segment_lengths widget (observed: all 4393), and normalStartFrame == start_frame. Change the timeline length and you must update all of them.
  3. Frames, not seconds. start / length are pixel-space frames and may be fractional (241.68). duration_seconds / frame_rate / time_units are display concerns.
  4. global_prompt exists twice, inside timeline_data and as a forceInput socket on the node. In the reference workflow the socket is wired from a PrimitiveStringMultiline; prefer the wired source and keep them consistent.
  5. guide_data must go somewhere. LTXDirector.guide_data → LTXDirectorGuide. That partner node is what injects the keyframes.
  6. Unsupported by the pack. The widget tooltip says "auto-managed; do not edit by hand". Well-formed blobs load fine; malformed JSON will break the editor. Round-trip and re-read after writing, and tell the user this is unofficial.

Reference wiring (from a working LTX 2.3 Director graph)

Power Lora Loader (rgthree) ─ model ─┐
DualCLIPLoader ──────────── clip ────┤
VAELoaderKJ ───────────── audio_vae ─┤   LTXDirector #42
PrimitiveStringMultiline ─ global_prompt ─┘
        │ model         → LTX2_NAG
        │ positive      → ConditioningZeroOut, LTXVConditioning
        │ audio_latent  → LTXVConcatAVLatent
        │ guide_data    → LTXDirectorGuide   ← required partner
        │ frame_rate    → AudioToFrameCount, VHS_VideoCombine
        └ combined_audio→ AudioToFrameCount, PreviewAudio, …

Procedure

  1. Find the node: get_workflow (action:"query") with types: ["LTXDirector"], fields: "detail" (or panel_query_graph on the live canvas).
  2. Read the current timeline_data and JSON.parse it. Never hand-splice the string.
  3. Mutate the parsed object (append a segment, flip a track gate, retime).
  4. Keep normalStartFrame / normalDurationFrames in sync with the widgets (edge 2).
  5. Write it back as a string via set_input.
  6. Re-read to confirm it round-trips, then create_workflow (action:"validate").

PromptRelayEncodeTimeline (pack: ComfyUI-PromptRelay) carries the same timeline_data hidden-widget pattern, so this skill's approach applies there too.

Any node whose controls are DOM widgets over hidden state is invisible to clicking and must be driven through its state widget. LTX Director is the clearest example.

Sources

  • Official: none found.
  • Empirical: timeline_data schema and edge rules from a working LTX 2.3 Director graph.

Similar Skills

brand-guidelines
anthropics/skills180k

brand-guidelines

Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.

AI & agents

internal-comms
anthropics/skills180k

internal-comms

A set of resources to help me write all kinds of internal communications, using the formats that my company likes to use. Claude should use this skill whenever asked to write some sort of internal communications (status reports, leadership updates, 3P updates, company newsletters, FAQs, incident reports, project updates, etc.).

AI & agents

template-skill
anthropics/skills180k

template-skill

Replace with description of the skill and when Claude should use it.

AI & agents

mcp-builder
anthropics/skills180k

mcp-builder

Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).

AI & agents

algorithmic-art
anthropics/skills180k

algorithmic-art

Creating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems. Create original algorithmic art rather than copying existing artists' work to avoid copyright violations.

AI & agents

academy-guide
anthropics/skills180k

academy-guide

Stop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content.

AI & agents