# Source notes — LTX Director in ComfyUI: Detailed LTX 2.3 Timeline, Prompt Relay, Keyframe, and Audio Guide

Retrieved: 2026-05-22

## Primary video
- URL: https://www.youtube.com/watch?v=vM60pJJqqEI
- Title: How to use LTX Director - A Free Open Source Tool for Creating LTX 2.3 AI Videos Locally in ComfyUI
- Channel: What Dreams Cost
- Channel URL: https://www.youtube.com/channel/UCTjTAamBpQKshIpoZjrKvlg
- Duration: 12:33
- Published/upload date from metadata: 2026-05-18

## Links from video description
- Node/repo: https://github.com/WhatDreamsCost/WhatDreamsCost-ComfyUI
- Example workflows: https://github.com/WhatDreamsCost/WhatDreamsCost-ComfyUI/tree/main/example_workflows
- LTX 2.3 prompting guide: https://ltx.io/model/model-blog/ltx-2-3-prompt-guide

## Verification notes
- GitHub README currently says WhatDreamsCost-ComfyUI can be installed by cloning into ComfyUI/custom_nodes or through ComfyUI Manager.
- README currently says if the latest version is not visible in Manager, download the nightly version or fetch updates; it also says ComfyUI-LTXVideo and ComfyUI-KJNodes must be updated, and ComfyUI-LTXVideo is required.
- pyproject.toml version observed: 1.3.9.
- GitHub example_workflows directory contains both full laid-out and subgraph LTX Director workflow variants.

## Transcript-derived procedure highlights
- Install What Dreams Cost via ComfyUI Manager, choosing nightly/latest when needed.
- Update ComfyUI LTX video nodes and ComfyUI KJ nodes.
- Load LTX Director templates from ComfyUI templates search or import from the repo.
- Use duration seconds/frames, frame rate, custom width/height, drag/drop image/audio, settings display mode, global prompt, custom audio toggle, timeline resizing, playback/zoom controls, segment prompt area, and guide strength.
- Prompt relay lets the user time specific prompt events across a clip rather than hoping the model performs them somewhere in the duration.
- Prompt failures usually come from prompts not written in a way LTX understands or from cramming too much action into too-short segments.
- Keyframes are guide images placed anywhere on the timeline; moving an image later can make the model generate what happens before it.
- Custom audio can drive motion/lip sync when custom audio is toggled on; for lip sync, describe the person speaking and optionally include the exact spoken words.
- Text-to-video uses text segments and custom width/height; global prompts apply atmosphere or persistent elements to the whole video.