<h1 id="wan2.1-infinitetalk-lipsync-workflow-guide"><strong>Wan2.1 InfiniteTalk LipSync — Workflow Guide</strong></h1><p><strong>52 nodes · 4 groups · 3 component subgraphs · 1 pipeline loop</strong> <strong>27 unique node types — 73% Eclipse nodes</strong> Built with <a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_Eclipse"><strong>ComfyUI_Eclipse</strong></a> custom nodes</p><hr /><h2 id="what-is-this"><strong>What Is This?</strong></h2><p>This workflow is a template designed to generate lip-synced talking head videos of arbitrary length using the <strong>Wan2.1 InfiniteTalk / LipSync</strong> model in ComfyUI.</p><p>The core feature is <strong>smart audio budgeting and seamless looping</strong>. The workflow analyzes the duration of a background audio track (speech), automatically computes how many generation loops are needed to match the track, recursively generates matching video blocks using temporal context, and blends them into a continuous video. It also includes a manual override switch to cap generations at a fixed loop count.</p><hr /><h2 id="how-it-works-the-basics"><strong>How It Works — The Basics</strong></h2><h3 id="wireless-data-routing-(setget)"><strong>Wireless Data Routing (Set/Get)</strong></h3><p>Rather than messy spaghetti wires running across the canvas, the workflow uses <strong>Set/Get nodes</strong> to route model, latent, audio, and loop count values. This keeps the layout clean and modular:</p><ul><li><p>A setter publishes a value (e.g. <code>Set_loop_count_ls</code> publishing <code>loop_count_ls</code>).</p></li><li><p>Getters retrieve the value by name wherever it is needed in the samplers and loop groups.</p></li></ul><h3 id="dual-switch-loop-control"><strong>Dual-Switch Loop Control</strong></h3><p>The loop iterations are controlled via the <strong>Any Dual-Switch [Eclipse]</strong> (id: 35) node. It allows the user to switch between:</p><ol><li><p><strong>Choice 1 (Manual)</strong>: Uses a static <code>loop_count</code> setting configured inside the <strong>Settings</strong> group panel.</p></li><li><p><strong>Choice 2 (Auto-Calculated)</strong>: Uses the dynamic loop count computed from the audio track's duration. By default, the workflow is configured to use the auto-calculated loops to automatically match the audio's length.</p></li></ol><h3 id="the-recursive-generation-loop"><strong>The Recursive Generation Loop</strong></h3><p>The generation process is structured inside an <code>easy forLoopStart</code> (id: 64) and <code>easy forLoopEnd</code> (id: 97) loop block:</p><ul><li><p><strong>Iteration 0 (Base Sampler)</strong>: The first block runs the <strong>Base Sampler</strong> group. It takes the initial start image (face) and the first segment of encoded audio to generate the beginning of the video.</p></li><li><p><strong>Iteration 1+ (Extend Sampler)</strong>: For subsequent loops, the <strong>Extend Sampler</strong> group runs. It takes the ending frames of the previous loop (<code>previous_frames</code>) as context to guide the model's starting state (ensuring visual continuity) and samples the next segment of speech audio. It outputs only the unique new frames (<code>trim_image</code>).</p></li></ul><hr /><h2 id="group-by-group-reference"><strong>Group-by-Group Reference</strong></h2><h3 id="1.-settings-(group-node)"><strong>1. Settings (Group Node)</strong></h3><p>This is the central configuration panel. It exposes:</p><ul><li><p><strong>Video Size &amp; Resolution</strong>: Sets output width and height (typically 480p or 720p).</p></li><li><p><strong>Frame Rate</strong>: Target output framerate (e.g. <code>24.0</code> or <code>30.0</code>).</p></li><li><p><strong>Manual Loop Override</strong>: A <code>loop_count</code> input to limit the loops when manual override is selected in the Dual-Switch.</p></li></ul><h3 id="2.-model-loaders"><strong>2. Model Loaders</strong></h3><ul><li><p><strong>Smart Model Loader v2 [Eclipse]</strong>: Loads the main Wan2.1 checkpoint, Text Encoder, and VAE. Default checkpoint is <code>Wan2_1-I2V-14B-720p_fp8_e4m3fn_scaled_KJ.safetensors</code> using the <code>default</code> template.</p></li><li><p><strong>Audio Encoder Loader &amp; Encode</strong>: Loads the speech analysis model <code>wav2vec2-chinese-base_fp16.safetensors</code> and encodes loaded speech audio into phonemic feature representations.</p></li><li><p><strong>Model Patch Loader</strong>: Applies the <code>wan2.1_infiniteTalk_multi_fp16.safetensors</code> patch to the diffusion model, adapting it for infinite talking generation.</p></li></ul><h3 id="3.-base-sampler-(group-node)"><strong>3. Base Sampler (Group Node)</strong></h3><p>A component subgraph containing 19 internal nodes:</p><ul><li><p>Uses <code>WanInfiniteTalkToVideo</code> to condition the initial start image and the first segment of the audio encoder output.</p></li><li><p>Utilizes a custom advanced sampler to generate the first talking head video block.</p></li></ul><h3 id="4.-extend-sampler-(group-node)"><strong>4. Extend Sampler (Group Node)</strong></h3><p>A component subgraph containing 19 internal nodes:</p><ul><li><p>Inherits the main model, conditioning, and audio encoder outputs.</p></li><li><p>Takes <code>previous_frames</code> (from the accumulated loop history) to guide the start of the next segment.</p></li><li><p>Generates and outputs <code>trim_image</code> (the newly generated frames with overlap cut off).</p></li></ul><h3 id="5.-loop-control-and-save-video"><strong>5. Loop Control &amp; Save Video</strong></h3><ul><li><p><strong>Image Join &amp; Loop Feedback</strong>: An <code>ImageBatch</code> node appends the newly generated frames from <code>Extend Sampler</code> to the accumulated video batch (<code>value1</code>), which is updated in the loop feedback loop.</p></li><li><p><strong>Save Video [Eclipse]</strong>: Takes the final accumulated image batch, remuxes the original audio file, and outputs an MP4. The <code>trim_mode</code> is set to <code>shortest</code>, which trims both the audio and video to the shorter of the two to guarantee perfect synchronization.</p></li></ul><hr /><h2 id="quick-start-guide"><strong>Quick Start Guide</strong></h2><h3 id="automatic-audio-budgeted-generation"><strong>Automatic Audio-budgeted Generation</strong></h3><ol><li><p>Verify that <strong>Any Dual-Switch [Eclipse]</strong> is set to <code>2</code> (Auto-calculated loops).</p></li><li><p>Load a face image in the <strong>Load Image</strong> node.</p></li><li><p>Load a voice clip in the <strong>Load Audio</strong> node.</p></li><li><p>Queue the prompt. The workflow will automatically compute the required loops, generate the segments, and output a perfectly timed talking head video.</p></li></ol><h3 id="manual-loop-count-generation"><strong>Manual Loop Count Generation</strong></h3><ol><li><p>Locate the <strong>Any Dual-Switch [Eclipse]</strong> (id: 35) and set its widget value to <code>1</code>.</p></li><li><p>Set your desired loop count in the <strong>Settings</strong> panel (under <code>loop_count</code>).</p></li><li><p>Queue the prompt. The generation will stop at your configured loop limit, regardless of how long the audio track is.</p></li></ol><h3 id="model-storage-locations-(for-local-users)"><strong>Model Storage Locations (for Local Users)</strong></h3><p>Ensure your model files are placed in these folders under your ComfyUI directory:</p><pre><code>📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └─── wan/Wan2_1-I2V-14B-720p_fp8_e4m3fn_scaled_KJ.safetensors
│   ├── 📂 text_encoders/
│   │   └─── nsfw_wan_umt5-xxl_bf16_fixed.safetensors
│   ├── 📂 model_patches/
│   │   ├─── wan2.1_infiniteTalk_single_fp16.safetensors
│   │   └─── wan2.1_infiniteTalk_multi_fp16.safetensors
│   ├── 📂 audio_encoders/
│   │   └─── wav2vec2-chinese-base_fp16.safetensors
│   └── 📂 vae/
│       └─── Wan2_1_VAE_bf16.safetensors
</code></pre><hr /><h2 id="custom-node-packages-used"><strong>Custom Node Packages Used</strong></h2><ul><li><p><a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_Eclipse">ComfyUI_Eclipse</a> — Custom loader templates, Set/Get wireless routing, Loop Calculators, and the Save/Preview Video nodes.</p></li><li><p><a target="_blank" rel="ugc" href="https://github.com/yolain/ComfyUI-Easy-Use">ComfyUI-Easy-Use</a> — The <code>easy forLoopStart</code> and <code>easy forLoopEnd</code> nodes for graph-level iteration.</p></li><li><p><a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a> — General utilities and crop helpers.</p></li></ul><p></p>

Files with SHA256: 090d04a9eac9fab682a93bab60159e2fce7a680b07ef6fd5cbfb7e4ff977a678