<h1 id="wan2.1-infinitetalk-lipsync-workflow-guide">Wan2.1 InfiniteTalk LipSync — Workflow Guide</h1>52 nodes · 4 groups · 3 component subgraphs · 1 pipeline loop 27 unique node types — 73% Eclipse nodes Built with <a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_Eclipse">ComfyUI_Eclipse</a> and <a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_SmartModelLoader">ComfyUI_Smart Model Loader</a> custom nodes<hr /><h2 id="what-is-this">What Is This?</h2>This workflow is a template designed to generate lip-synced talking head videos of arbitrary length using the Wan2.1 InfiniteTalk / LipSync model in ComfyUI.The core feature is smart audio budgeting and seamless looping. 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.<hr /><h2 id="how-it-works-the-basics">How It Works — The Basics</h2><h3 id="wireless-data-routing-(setget)">Wireless Data Routing (Set/Get)</h3>Rather than messy spaghetti wires running across the canvas, the workflow uses Set/Get nodes to route model, latent, audio, and loop count values. This keeps the layout clean and modular:<ul><li>A setter publishes a value (e.g. <code>Set_loop_count_ls</code> publishing <code>loop_count_ls</code>).</li><li>Getters retrieve the value by name wherever it is needed in the samplers and loop groups.</li></ul><h3 id="dual-switch-loop-control">Dual-Switch Loop Control</h3>The loop iterations are controlled via the Any Dual-Switch [Eclipse] (id: 35) node. It allows the user to switch between:<ol><li>Choice 1 (Manual): Uses a static <code>loop_count</code> setting configured inside the Settings group panel.</li><li>Choice 2 (Auto-Calculated): 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.</li></ol><h3 id="the-recursive-generation-loop">The Recursive Generation Loop</h3>The generation process is structured inside an <code>easy forLoopStart</code> (id: 64) and <code>easy forLoopEnd</code> (id: 97) loop block:<ul><li>Iteration 0 (Base Sampler): The first block runs the Base Sampler group. It takes the initial start image (face) and the first segment of encoded audio to generate the beginning of the video.</li><li>Iteration 1+ (Extend Sampler): For subsequent loops, the Extend Sampler 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>).</li></ul><hr /><h2 id="group-by-group-reference">Group-by-Group Reference</h2><h3 id="1.-settings-(group-node)">1. Settings (Group Node)</h3>This is the central configuration panel. It exposes:<ul><li>Video Size &amp; Resolution: Sets output width and height (typically 480p or 720p).</li><li>Frame Rate: Target output framerate (e.g. <code>24.0</code> or <code>30.0</code>).</li><li>Manual Loop Override: A <code>loop_count</code> input to limit the loops when manual override is selected in the Dual-Switch.</li></ul><h3 id="2.-model-loaders">2. Model Loaders</h3><ul><li>Smart Model Loader v2 [Eclipse]: 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.</li><li>Audio Encoder Loader &amp; Encode: Loads the speech analysis model <code>wav2vec2-chinese-base_fp16.safetensors</code> and encodes loaded speech audio into phonemic feature representations.</li><li>Model Patch Loader: Applies the <code>wan2.1_infiniteTalk_multi_fp16.safetensors</code> patch to the diffusion model, adapting it for infinite talking generation.</li></ul><h3 id="3.-base-sampler-(group-node)">3. Base Sampler (Group Node)</h3>A component subgraph containing 19 internal nodes:<ul><li>Uses <code>WanInfiniteTalkToVideo</code> to condition the initial start image and the first segment of the audio encoder output.</li><li>Utilizes a custom advanced sampler to generate the first talking head video block.</li></ul><h3 id="4.-extend-sampler-(group-node)">4. Extend Sampler (Group Node)</h3>A component subgraph containing 19 internal nodes:<ul><li>Inherits the main model, conditioning, and audio encoder outputs.</li><li>Takes <code>previous_frames</code> (from the accumulated loop history) to guide the start of the next segment.</li><li>Generates and outputs <code>trim_image</code> (the newly generated frames with overlap cut off).</li></ul><h3 id="5.-loop-control-and-save-video">5. Loop Control &amp; Save Video</h3><ul><li>Image Join &amp; Loop Feedback: 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.</li><li>Save Video [Eclipse]: 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.</li></ul><hr /><h2 id="quick-start-guide">Quick Start Guide</h2><h3 id="automatic-audio-budgeted-generation">Automatic Audio-budgeted Generation</h3><ol><li>Verify that Any Dual-Switch [Eclipse] is set to <code>2</code> (Auto-calculated loops).</li><li>Load a face image in the Load Image node.</li><li>Load a voice clip in the Load Audio node.</li><li>Queue the prompt. The workflow will automatically compute the required loops, generate the segments, and output a perfectly timed talking head video.</li></ol><h3 id="manual-loop-count-generation">Manual Loop Count Generation</h3><ol><li>Locate the Any Dual-Switch [Eclipse] (id: 35) and set its widget value to <code>1</code>.</li><li>Set your desired loop count in the Settings panel (under <code>loop_count</code>).</li><li>Queue the prompt. The generation will stop at your configured loop limit, regardless of how long the audio track is.</li></ol><h3 id="model-storage-locations-(for-local-users)">Model Storage Locations (for Local Users)</h3>Ensure your model files are placed in these folders under your ComfyUI directory:<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">Custom Node Packages Used</h2><ul><li><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.</li><li><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.</li><li><a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a> — General utilities and crop helpers.</li></ul>

Wan2.1 InfiniteTalk LipSync (Native)

<ul><li>requires <a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_Eclipse">ecipse 4.3.20</a></li><li>updated the subgraphs to work with 1.48.7 / comfyui 0.33.1</li><li>added the new video save with gen data node (reading the metadata is not supported by civitai right now... pending)</li></ul>

Files with SHA256: ca4f0ccdd0821baca8fa1307f8bb8dbab223f5c267f2b069484d0bf8c760b8db