Lightricks/ComfyUI-LTXVideo

What it solves

This project provides a set of custom nodes and workflows for ComfyUI to fully leverage the LTX-2 video generation model. It expands the core ComfyUI LTX-2 integration by adding advanced control mechanisms, high-dynamic-range (HDR) output, audio generation, and creative upscaling tools.

How it works

It integrates several specialized LoRAs (Low-Rank Adaptation) and custom nodes into the ComfyUI graph-based interface. These include:

• Union IC-LoRA: A unified model that handles both depth and edge (Canny) control signals simultaneously using downsampled latents for faster inference.
• HDR IC-LoRA: Generates video in ARRI LogC3 space, which is then decoded into linear HDR values for professional grading and EXR export.
• Dub-It IC-LoRA: A multilingual dubbing tool that regenerates lip movements and audio to match a target text prompt while preserving the speaker's identity.
• Pixel Spatial Upscaler: A generative upscaler that synthesizes new fine details (2x or 4x) rather than just interpolating pixels.
• Audio-Only Mode: Utilizes the joint audio/video transformer to generate audio from text without producing video.

Who it’s for

Content creators, VFX artists, and AI video researchers who use ComfyUI and have access to high-VRAM GPUs (32GB+ recommended) to generate controlled, high-quality video and audio content.

Highlights

• Unified Control: Combines depth and edge control into a single LoRA.
• Professional HDR: Supports linear HDR output and 16/32-bit EXR image sequences.
• AI Dubbing: Multilingual speech rephrasing and lip-syncing.
• Generative Upscaling: Creative spatial upscaling that adds synthesized texture and structure.
• Joint Audio/Video: Ability to perform text-to-audio generation independently.