comfyui-ltx2-efficient

LTX2 Efficient Video Sampler

ComfyUI License

A high-performance custom node for LTX2 video generation in ComfyUI, designed to drastically reduce GPU usage, VRAM requirements, and temperature on consumer GPUs (especially 6GB cards like RTX 2060).

GPU Temperature

✨ Features

🎯 Core Optimizations

• Frame-Stride Diffusion: Process only keyframes (e.g., 1 out of 4), interpolate the rest
• Spatial Attention Freezing: Skip redundant attention computation after configurable threshold
• LTX2-Specific Hooks: Precise targeting of attn1 (self-attention) vs attn2 (cross-attention)

🌡️ Thermal Management

• Real-Time GPU Monitoring: Uses pynvml to read actual GPU temperature
• Dynamic Throttling: Auto-adjusts processing speed to maintain target temperature
• Emergency Protection: Automatic 2-second pause at critical temps (83°C+)
• Preset Profiles: Quick optimization for different GPU tiers

🎬 Advanced Interpolation

• Linear: Fast, basic blending
• Slerp: Spherical interpolation - smoother for diffusion latents
• Motion-Compensated: Optical flow-based warping for best motion quality

📦 Additional Nodes

• LTX2 Temporal VAE Decode: Memory-efficient VAE decoding with temporal tiling
• LTX2 Separate Audio/Video Latent: Extract video and audio from combined latents (for use with LTXVConcatAVLatent)
• LTX2 Combine Audio/Video Latent: Recombine processed video with audio for downstream nodes
• LTX2 Empty Audio Latent: Generate placeholder audio latent matching video duration

🎵 Audio-Video Workflow

For workflows using LTXVConcatAVLatent or MMAudio:

[LTXVConcatAVLatent] → [LTX2 Separate AV] → [LTX2 Efficient Sampler] → [LTX2 Combine AV] → [Decode]
                                         ↓                              ↑
                                  [Audio Pipeline (optional)] ──────────┘

📋 Requirements

• ComfyUI (latest)
• PyTorch with CUDA
• pynvml (optional, for thermal monitoring)

🚀 Installation

Option 1: ComfyUI Manager (Recommended)

Search for "LTX2 Efficient" in ComfyUI Manager and install.

Option 2: Manual Installation

cd ComfyUI/custom_nodes/
git clone https://github.com/kakachiex2/LTX2-Efficient-Sampler
cd comfyui-ltx2-efficient
pip install -r requirements.txt

🎛️ Node Settings

Optimization Presets

Preset	Freeze Ratio	Throttle	Frame Stride	Best For
Performance (RTX 3080+)	70%	0ms	2	High-end GPUs
Balanced (RTX 3060/3070)	50%	25ms	4	Mid-range
Aggressive (RTX 2060/2070)	30%	50ms	4	6GB VRAM
Ultra Low Power (GTX 1660)	15%	100ms	6	Older GPUs
Extreme Cool (RTX 2060 6GB)	10%	150ms	8	Target 65°C
Thermal Auto-Scale	30%	Dynamic	4	Recommended
Custom	Manual	Manual	Manual	Fine-tuning

Key Parameters

Parameter	Default	Description
optimization_preset	Thermal Auto-Scale	Quick optimization profile
target_temp	70°C	Target GPU temperature (for auto-scale)
frame_stride	4	Process every Nth frame (higher = faster)
freeze_ratio	0.3	Start freezing attention at this % o