FL LTX Tools

Experimental tools and motion controls for LTX-Video in ComfyUI. Designed to drop into existing LTX workflows with minimal wiring.

Features

• Single-knob motion control — one motion_intensity slider that drives multiple underlying LTX motion levers
• VAE decoder noise control — exposes decode_noise_scale and decode_timestep for per-frame stochasticity
• Drop-in replacement for the built-in LTXVConditioning node — same conditioning pattern, with motion boost on top
• Self-contained — no dependency on other LTX custom node packs

Nodes

Node	Description
FL LTX Motion Boost	Drives motion via conditioning frame_rate (RoPE temporal scale) + VAE decoder noise. Inputs: positive/negative cond + VAE. Outputs: patched cond + VAE.

Installation

ComfyUI Manager

Search for "FL LTX Tools" and install.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI-FL-LTXTools.git
cd ComfyUI-FL-LTXTools
pip install -r requirements.txt

Quick Start

1. Build a normal LTX-Video workflow (model loader, text encode, empty latent, sampler, VAE decode)
2. Insert FL LTX Motion Boost between your text encoders and the sampler
3. Wire positive, negative, and vae through it
4. Adjust motion_intensity (0.0 = static, 0.5 = balanced, 1.0 = chaotic)

How It Works

LTX-Video has no single "motion intensity" knob — motion behavior emerges from the interaction of multiple parameters. This node consolidates the most reliable motion lever (conditioning frame_rate via RoPE temporal embeddings) with VAE-side stochasticity into a single intuitive control.

motion_intensity

Maps to the conditioning frame_rate value:

• 0.0 → uses base_frame_rate (default 25 fps) — no motion boost
• 0.5 → ~17 fps — moderate boost (LTX interprets each frame as a larger temporal step)
• 1.0 → ~10 fps — heavy boost, dramatic motion per frame

Lower frame_rate values widen the gaps between RoPE temporal positions, so the model "thinks" each output frame represents a longer time slice. Result: more motion happens between frames.

decode_noise + decode_timestep

VAE decoder noise injection. Adds Gaussian stochasticity at decode time:

• Higher noise → smoother, more organic motion textures
• Lower noise → crisper but potentially jittery output
• LTX defaults: decode_noise = 0.025, decode_timestep = 0.05

Inputs

Name	Type	Range	Default	Description
positive	CONDITIONING	—	—	Positive conditioning
negative	CONDITIONING	—	—	Negative conditioning
vae	VAE	—	—	LTX VAE
motion_intensity	FLOAT	0.0–1.0	0.5	Motion boost amount
decode_noise	FLOAT	0.0–0.5	0.025	VAE decode noise scale
decode_timestep	FLOAT	0.0–0.5	0.05	VAE decode noise timestep
base_frame_rate	INT (optional)	6–60	25	Reference fps before scaling
override_frame_rate	INT (optional)	-1, 6–60	-1	Set explicit fps, ignore motion_intensity

Outputs

(positive, negative, vae) — drop into your sampler / VAE decode chain.

Roadmap

More LTX experiments planned. Potential additions:

• STG (Spatiotemporal Skip Guidance) helper
• Motion preset library (cinematic, action, dialogue, etc.)
• Latent noise manipulation utilities

License

Apache-2.0 — Built on top of Lightricks LTX-Video.