The LTX 2.3 KJ Stripped 1.1 workflow is a specialized audio-visual (AV) generation pipeline designed to interpolate between two static images. It utilizes a "Guided" architecture to ensure the generated video adheres strictly to a designated start and end point.<h3 id="core-architecture-and-processing">Core Architecture &amp; Processing</h3><ul><li>Model Loading: The workflow employs the LTX-Video 2.3 22B Distilled transformer model via a specialized KJ loader node.</li><li>Dual-Modality Processing: It initializes both an <code>EmptyLTXVLatentVideo</code> and an <code>LTXVEmptyLatentAudio</code> space.</li><li>AV Concatenation: Visual and audio latents are merged into a single latent stream using the <code>LTXVConcatAVLatent</code> node, allowing the sampler to process both simultaneously.</li></ul><h3 id="frame-guidance-system">Frame Guidance System</h3><ul><li>Start Frame Anchor: An image is loaded and preprocessed to serve as the reference for <code>frame_idx: 0</code>.</li><li>End Frame Anchor: A second image is loaded and preprocessed to serve as the reference for the final frame (<code>frame_idx: -1</code>).</li><li>Guide Application: The <code>LTXVAddGuide</code> nodes apply these images to the latent space with a configurable strength (set to 0.7 in this version) to dictate the video’s trajectory.</li></ul><h3 id="sampling-and-decoding">Sampling &amp; Decoding</h3><ul><li>KSampler Configuration: The workflow uses a high-speed 8-step Euler sampler with a simple scheduler.</li><li>Separation &amp; Decoding: After sampling, the <code>LTXVSeparateAVLatent</code> node splits the results back into distinct video and audio streams.</li><li>Tiled Decoding: Visuals are decoded using <code>VAEDecodeTiled</code> to manage high-resolution output (1024x1024) efficiently.</li></ul><h3 id="enhancement-and-output">Enhancement &amp; Output</h3><ul><li>NVIDIA RTX Integration: The workflow includes an <code>RTXVideoSuperResolution</code> node that provides a 2x hardware-accelerated upscale on the final image sequence.</li><li>Final Assembly: The <code>VHS_VideoCombine</code> nodes produce two versions of the video—a base generation and a super-resolved version—both featuring synchronized, AI-generated audio.</li></ul>

LTX 2.3 FLF2V

ltx23FLF2V_ltx23KJStripped11.zip

LTX 2.3 KJ Stripped 1.1

RTX Upscale Version

v1.0

A comprehensive production-grade pipeline designed for the LTX-2 model. It specializes in generating high-fidelity video by combining a source image and an audio track to create synchronized content, such as music videos with lip-syncing and dance.<h3 id="key-features-and-architecture">Key Features &amp; Architecture</h3>The workflow is organized into distinct logical stages using subgraphs to manage complexity and optimize hardware resources:<ul><li>Multimodal Input Processing:<ul><li>Image Handling: Uses <code>ImageResizeKJv2</code> to prepare a source image, which acts as the visual foundation for the video.</li><li>Audio Integration: Employs a <code>VHS_LoadAudioUpload</code> node to bring in external audio files, which guide the timing and motion of the generation.</li></ul></li><li>Dual-Stage Sampling Pipeline:<ul><li>Stage 1 (Initial Generation): Focuses on establishing the core motion and structure.</li><li>Stage 2 (Refinement): A secondary pass that refines the video and audio latents for higher quality.</li></ul></li><li>VRAM Optimization:<ul><li>Gemma API Text Encode: Instead of loading the massive Gemma-3 12B model locally, this workflow uses an API-based text encoder. This significantly reduces local VRAM requirements, allowing the workflow to run on GPUs with as little as 12GB to 16GB.</li></ul></li><li>Creative Controls:<ul><li>Camera LoRAs: Includes dedicated slots for LTX-2 Camera Control LoRAs (e.g., Dolly Left), allowing for precise cinematic movement.</li><li>Latent Upscaling: Incorporates a spatial upscaler to enhance the resolution of the final output.</li></ul></li></ul>

Files with SHA256: 580f79a076d0393debc4fd12e318c84b3b66ccbf2589e92eddec30b2706f4265