LTXVLoopingSampler
The LTXVLoopingSampler addresses two fundamental challenges in video generation: 1. Temporal Length: Generating very long videos by dividing them into overlapping temporal segments 2. Spatial Resolution: Creating high-resolution frames by splitting them into overlapping spatial regions Each "tile" is processed independently and then seamlessly blended with its neighbors to create the final coherent video output. - 🎬 Temporal Tiling: Generates long videos by processing overlapping time segments - 🖼️ Spatial Tiling: Creates high-resolution output by dividing frames into spatial regions - 🎯 Multiple Conditioning Modes: Supports image conditioning, guiding latents, and negative index conditioning - ⚡ Memory Efficient: Processes one spatial tile at a time to minimize memory usage - 🔄 Seamless Blending: Uses weighted blending for smooth transitions between tiles - 🎲 Advanced Seeding: Configurable per-tile seeding for reproducible results | Parameter | Type | Default | Range | Description | |-----------|------|---------|-------|-------------| | model | MODEL | - | - | The diffusion model for video generation | | vae | VAE | - | - | VAE for encoding/decoding between pixel and latent space | | noise | NOISE | - | - | Noise generator for the sampling process | | sampler | SAMPLER | - | - | Sampling algorithm (e.g., Euler, DPM++) | | sigmas | SIGMAS | - | - | Noise schedule defining the denoising steps | | guider | GUIDER | - | - | Conditioning guider (must be STGGuiderAdvanced to work with LTXVLoopingSampler) | | latents | LATENT | - | - | Input latent tensor defining video dim…
Sampling · 12 Eingaben · 12 Parameter · 1 Ausgabe
