Built by Muse Collective — generate long-form AI video with lipsync, custom audio, background ambience, and per-segment prompts from a single node with a built-in visual timeline editor.

This repo: V2 with Seed Hunt

This is a fork of the original muse-collective-26/muse-ltx-timeline (V1), which is included here untouched. This repo adds a second node, Muse Collective LTX Timeline V2, with one new feature: Seed Hunt.

LTX 2.3 is far more sensitive to seed than to prompt wording — the fastest path to a good result is usually trying a handful of seeds cheaply and picking the best one, rather than iterating on the prompt. Seed Hunt bakes that workflow into the node itself, using the node's own real timeline data (images, audio, motion guide) so multi-segment timelines just work with no extra wiring.

How it works:

Flip seed_hunt on, leave all four use_seed_hunt_1..4 toggles off, and run — the node generates 4 candidates at Stage 1's real resolution, each with a fresh random seed, instead of running the full pipeline. Output ports seed_hunt_preview_1..4 show each candidate, and each candidate's actual Stage 1 latent is cached in memory.
Look at the 4 previews, then flip exactly one use_seed_hunt_N on and run again — the node skips regenerating Stage 1 from scratch and instead carries that candidate's cached real latent straight into the existing Stage 2 step (upscale + partial refine). What you see in the preview is what carries through, not a fresh independent roll that merely happens to share a seed number.
If the cache is empty (e.g. the server restarted since scouting, or you commit to a candidate without ever scouting first), it falls back to overriding seed and running Stage 1 fresh, using that candidate's seed_hunt_N widget value.

This mirrors the community "ltx23SeedHunter" workflow that inspired this feature: the seed number doesn't need to "match" across resolutions, because Stage 2 is refining the actual picked content, not regenerating it from a noise field that merely shares a seed integer.

Settings added on top of V1:

| Setting | Purpose | |---|---| | seed_hunt | Master toggle for the feature. Off = identical to V1. | | use_seed_hunt_1..4 | Flip exactly one on to commit to that candidate. | | seed_hunt_steps | Sampler steps for the scouting pass (independent of stage1_steps). Advanced — hidden behind the Settings toggle by default. | | seed_hunt_1..4 | Fallback seeds, only used if you commit to a candidate without ever scouting (cache miss) — normal scouting ignores these and randomizes instead. Advanced — hidden behind the Settings toggle by default. | | seed_hunt_scale | Unused as of 1.0.4 — scouting always renders at Stage 1's real resolution now. Kept only so older saved workflows still load correctly; permanently hidden in the UI. |

Also included: Muse Seed Scout, an earlier, standalone version of the same idea that runs outside Director and only takes a single reference image rather than the full timeline. Superseded by V2's built-in Seed Hunt for anything with more than one timeline segment, but kept as a lighter-weight option for single-image setups.

One deliberate difference from V1: V1 uses seed - 1 for Stage 2's noise seed (separate from Stage 1's); V2 uses the same seed value for both stages. That said, Seed Hunt's faithfulness to a picked candidate doesn't come from the seed matching — it comes from Stage 2 refining that candidate's actual cached latent directly.

What it does

A full timeline-based node for generating long-form AI video with LTX 2.3 AV — chunked generation, reference-frame continuity, per-segment prompts, lipsync, and layered audio — all from a single node.

Key features

Visual timeline editor — drag-and-drop images, audio, and video segments directly onto the timeline
Infinite chunking — generate 90 seconds+ by splitting into overlapping chunks with carry-frame latent locking for seamless transitions
Per-segment prompting — different prompts for different parts of the video using PromptRelay temporal attention masking
Lipsync — sync mouth movements to a custom speech audio file using the LTX talking head LoRA
Three audio layers — generated ambient audio, custom speech/music, and a separate BG ambience track
IC-LoRA support — reference video for camera motion or style consistency
Retake mode — replace a section of an existing video with a new generation
Dual-pass sampling — Stage 1 generation + Stage 2 spatial upscaler pass
Color matching — automatic color correction between chunks

Nodes included

| Node | Description | |------|-------------| | MuseDirectorSamplerV1 | Main director + infinite sampler with timeline UI | | MuseDirectorSamplerV2 | Fork of V1 with the Seed Hunt scouting feature (see above) | | MuseSeedScout | Standalone cheap 4-seed scouting node for single-image setups outside Director | | MuseGuide | Standalone reference keyframe and IC-LoRA guide encoder — wires directly into LTX conditioning | | MuseCropGuides | Trims guide keyframes to match a cropped or upscaled latent — use after the spatial upscaler | | LTXInfiniteDirectorSamplerV7 | Previous chunked sampler (no timeline UI) |

Requirements

ComfyUI custom nodes (install via Manager)

ComfyUI-VideoHelperSuite — required for video preview and saving

Python packages

pip install av torchaudio soundfile

Installation

Via ComfyUI Manager

Search for Muse Collective LTX Timeline and click Install.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/muse-collective-26/muse-ltx-timeline

Restart ComfyUI after installing.

Model setup

Download all models and place them in the correct subfolders inside your ComfyUI models/ directory.

Text Encoder (CLIP)

Download: gemma_3_12B_it_fp4_mixed.safetensors Place in: models/text_encoders/

Spatial Upscaler

Download: ltx-2.3-spatial-upscaler-x2-1.1.safetensors Place in: models/latent_upscale_models/

Talking Head LoRA (required for Lipsync)

Download: LTX-2.3-22b-AV-LoRA-talking-head-v1.safetensors Place in: models/loras/LTX2.3/

Example workflow

An example workflow is included in the repo: Muse-Director-V2 - seed scout.json

This workflow uses MuseDirectorSamplerV2 at 1920×1088, and includes:

All 4 seed_hunt_preview_N / seed_hunt_audio_N outputs wired to their own Video Combine nodes for reviewing Seed Hunt candidates before committing
Lipsync-ready audio wiring
The reference_image output wired into MuseColorMatchBatched (a separate, memory-bounded color-match node — muse-collective-batched-colour-match, installed independently), which feeds both a direct "Full" output and an RTX Video Super Resolution "Upscaled" output

Load it in ComfyUI via Load and swap in your own image and audio files.

Node connections

| Input | Type | Description | |-------|------|-------------| | model | MODEL | LTX model with LoRA applied (use talking head LoRA if using lipsync) | | clip | CLIP | LTX text encoder | | audio_vae | VAE | LTX Audio VAE | | vae | VAE | LTX Video VAE | | spatial_upscaler | LATENT_UPSCALE_MODEL | LTX spatial upscaler | | bg_audio | AUDIO (optional) | Background ambience — mixed under everything | | base_model | MODEL (optional) | Base model without LoRA — wire the UNETLoader output directly here for ambient audio generation without the talking head LoRA influencing it |

Timeline tracks

| Track | Purpose | |-------|---------| | MAIN | Image/video segments — the reference frame(s) for generation | | AUDIO | Speech or music clips for Custom Audio mode | | BG AUDIO | Background ambience — layered under everything, volume controlled by bg_volume | | MOTION | Reference video segments for IC-LoRA motion guidance |

Audio toggles

Three buttons in the node toolbar control audio behaviour. They can be used individually or in combination.

Gen Audio (ON only)

LTX generates ambient sound and atmosphere from [SOUNDS] tags in your prompts. No audio file needed — the model creates sound that matches the scene description.

[SOUNDS] Busy coffee shop, cups clinking, espresso machine, background chatter

Custom Audio (ON only)

Plays the audio file(s) from the AUDIO timeline track directly in the output. The audio is passed through as-is — no generation, no lipsync. Use this for background music, a pre-recorded voice-over, or any audio you want to play over the video without affecting the visuals.

Custom Audio + Lipsync (Custom Audio ON, Lipsync ON, Gen Audio OFF)

Drives the character's lip movements to match the speech in your custom audio file.

Requirements:

A speech audio file on the AUDIO timeline track
Custom Audio button ON
Lipsync button ON
Talking head LoRA loaded into the model input

The model will sync mouth movements to the speech. The original audio file plays in the output — it is not re-generated.

Gen Audio + Custom Audio (both ON, Lipsync OFF)

Generates ambient audio from [SOUNDS] prompts AND plays the custom audio file. Both are mixed together in the output. Use this for generated atmosphere layered under a pre-recorded music track or narration that does not require lipsync.

BG Audio track

The BG AUDIO track is independent of all three toggle buttons. Drop any audio file onto it and it will be mixed under the main audio output at the volume level set by bg_volume. Works with any combination of the other audio modes.

Prompt format

Use uppercase tags inside segment and global prompts:

A woman sits at a podcast desk, talking confidently to camera.
[SPEECH] Right. I'm going to tell you something most people in this space won't admit.
[SOUNDS] Quiet studio, soft air conditioning hum, distant city traffic

| Tag | Effect | |-----|--------| | [SPEECH] | The words the character should say — used with lipsync or generated audio | | [SOUNDS] | Ambient sounds and atmosphere — used when Gen Audio is ON |

Important: Tags must be uppercase. [speech] and [sounds] (lowercase) are ignored by the model.

Chunking settings

| Setting | Description | |---------|-------------| | chunk_duration_seconds | Length of each generated chunk (default 30s) | | auto_chunk_threshold | Videos longer than this are split into chunks automatically | | carry_frames | Frames locked from the previous chunk as a reference (default 73 ≈ 3s at 24fps) | | carry_strength | How strongly carry frames anchor the next chunk (default 1.0) | | crossfade_frames | Blend frames at chunk boundaries to smooth transitions |

Recommended settings

| Use case | Resolution | Chunk | Steps S1 / S2 | Denoise S2 | |----------|-----------|-------|---------------|------------| | Talking head 9:16 | 704×1280 | 30s | 8 / 4 | 0.42 | | Short clip 16:9 | 960×544 | 10s | 8 / 4 | 0.42 | | Long form 9:16 | 704×1280 | 30s | 8 / 4 | 0.42 |

Credits

Inspired by the LTX Director approach in WhatDreamsCost-ComfyUI by WhatDreamsCost — the MuseGuide and MuseCropGuides nodes are a standalone reimplementation that removes the WDC dependency entirely
LTX 2.3 AV model by Lightricks
Built and extended by Muse Collective

License

MIT — free to use, modify and distribute.