[{"data":1,"prerenderedAt":509},["ShallowReactive",2],{"content-/en/advanced-tutorial/wan2.2/":3},{"id":4,"title":5,"body":6,"description":501,"extension":502,"meta":503,"navigation":504,"path":505,"seo":506,"stem":507,"__hash__":508},"content/en/advanced-tutorial/wan2.2.md","Wan22",{"type":7,"value":8,"toc":477},"minimark",[9,22,25,30,37,42,82,84,88,91,165,167,171,174,178,185,207,211,214,257,259,263,266,270,350,352,356,359,363,414,416,420,424,435,439,442,468,470,474],[10,11,12,13,17,18,21],"p",{},"This guide provides a comprehensive breakdown of ",[14,15,16],"strong",{},"Wan2.2",", the latest evolution in open-source video generation from Alibaba’s Tongyi Lab. Whether you are a professional motion designer or a hobbyist, integrating Wan2.2 into ",[14,19,20],{},"ComfyUI"," offers unparalleled control over cinematic AI video production.",[23,24],"hr",{},[26,27,29],"h2",{"id":28},"_1-what-is-wan22-the-architecture-revolution","1. What is Wan2.2? The Architecture Revolution",[10,31,32,33,36],{},"Released in early 2026, Wan2.2 isn't just an incremental update to Wan2.1; it introduces a ",[14,34,35],{},"Mixture-of-Experts (MoE)"," architecture to the world of video diffusion.",[38,39,41],"h3",{"id":40},"key-technical-highlights","Key Technical Highlights:",[43,44,45,60,70,76],"ul",{},[46,47,48,51,52,55,56,59],"li",{},[14,49,50],{},"MoE Architecture:"," Unlike traditional \"dense\" models that activate all parameters for every calculation, Wan2.2 uses a dual-expert system. It features a ",[14,53,54],{},"High-Noise Expert"," for initial scene layout/motion and a ",[14,57,58],{},"Low-Noise Expert"," for fine-textured details.",[46,61,62,65,66,69],{},[14,63,64],{},"Efficiency:"," Despite having 27B total parameters, only ",[14,67,68],{},"14B are active"," at any given time. This allows for high-tier quality with the VRAM footprint of a much smaller model.",[46,71,72,75],{},[14,73,74],{},"Cinematic Aesthetics:"," The model was trained on a dataset with over 80% more video content than its predecessor, specifically labeled for lighting, contrast, and professional camera movement.",[46,77,78,81],{},[14,79,80],{},"Performance:"," Capable of native 720p and 1280p outputs at 24fps.",[23,83],{},[26,85,87],{"id":86},"_2-model-variants-14b-vs-5b","2. Model Variants: 14B vs. 5B",[10,89,90],{},"Choosing the right version depends entirely on your hardware:",[92,93,94,114],"table",{},[95,96,97],"thead",{},[98,99,100,105,108,111],"tr",{},[101,102,104],"th",{"align":103},"left","Model",[101,106,107],{"align":103},"Active Params",[101,109,110],{"align":103},"Recommended VRAM",[101,112,113],{"align":103},"Best Use Case",[115,116,117,134,149],"tbody",{},[98,118,119,125,128,131],{},[120,121,122],"td",{"align":103},[14,123,124],{},"Wan2.2-T2V-A14B",[120,126,127],{"align":103},"14B",[120,129,130],{"align":103},"24GB+ (RTX 3090/4090)",[120,132,133],{"align":103},"High-end cinematic Text-to-Video",[98,135,136,141,143,146],{},[120,137,138],{"align":103},[14,139,140],{},"Wan2.2-I2V-A14B",[120,142,127],{"align":103},[120,144,145],{"align":103},"24GB+",[120,147,148],{"align":103},"Professional Image-to-Video (consistent)",[98,150,151,156,159,162],{},[120,152,153],{"align":103},[14,154,155],{},"Wan2.2-TI2V-5B",[120,157,158],{"align":103},"5B",[120,160,161],{"align":103},"10GB - 12GB",[120,163,164],{"align":103},"Fast iterations on consumer GPUs",[23,166],{},[26,168,170],{"id":169},"_3-installation-guide-for-comfyui","3. Installation Guide for ComfyUI",[10,172,173],{},"To run Wan2.2, you need an up-to-date ComfyUI installation and the specific custom nodes designed for Wan's video wrappers.",[38,175,177],{"id":176},"step-1-install-custom-nodes","Step 1: Install Custom Nodes",[10,179,180,181,184],{},"Open your ",[14,182,183],{},"ComfyUI Manager"," and search for:",[186,187,188,195,201],"ol",{},[46,189,190,194],{},[191,192,193],"code",{},"ComfyUI-WanVideoWrapper"," (by Kijai): The most stable implementation for Wan2.2.",[46,196,197,200],{},[191,198,199],{},"ComfyUI-VideoHelperSuite",": Essential for loading images and saving MP4/GIF outputs.",[46,202,203,206],{},[191,204,205],{},"ComfyUI-KJNodes",": Provides specialized mask and noise tools.",[38,208,210],{"id":209},"step-2-download-the-weights","Step 2: Download the Weights",[10,212,213],{},"Place your model files in the following directories:",[43,215,216,230,239],{},[46,217,218,221,222,225,226,229],{},[14,219,220],{},"Diffusion Model:"," ",[191,223,224],{},"ComfyUI/models/checkpoints/"," (or ",[191,227,228],{},"models/diffusion_models/"," depending on your node version).",[46,231,232,221,235,238],{},[14,233,234],{},"VAE:",[191,236,237],{},"ComfyUI/models/vae/"," (Ensure you use the dedicated Wan2.2 VAE for proper 16x16x4 compression).",[46,240,241,244,245,248,249,252,253,256],{},[14,242,243],{},"Text Encoders:"," Usually requires ",[14,246,247],{},"T5-v1.1-xxl"," and ",[14,250,251],{},"UMT5",", placed in ",[191,254,255],{},"models/clip/",".",[23,258],{},[26,260,262],{"id":261},"_4-text-to-video-t2v-workflow-tutorial","4. Text-to-Video (T2V) Workflow Tutorial",[10,264,265],{},"Generating a video from scratch requires a structured prompt and proper sampler settings.",[38,267,269],{"id":268},"the-node-setup","The Node Setup",[186,271,272,282,296,311],{},[46,273,274,277,278,281],{},[14,275,276],{},"WanVideo Loader:"," Select the ",[191,279,280],{},"Wan2.2-T2V-14B"," checkpoint.",[46,283,284,287,288,291,292,295],{},[14,285,286],{},"Empty Wan Latent:"," Set your resolution. For 14B, ",[14,289,290],{},"1280x720"," is the sweet spot. Set frames to ",[14,293,294],{},"81"," (approx. 5 seconds at 24fps).",[46,297,298,301,302],{},[14,299,300],{},"CLIP Text Encode:"," Wan2.2 understands natural language better than tags.\n",[43,303,304],{},[46,305,306,310],{},[307,308,309],"em",{},"Good Prompt:"," \"A cinematic tracking shot of a futuristic cyberpunk city during a rainstorm, neon lights reflecting on puddles, hyper-realistic, 8k, high contrast.\"",[46,312,313,316],{},[14,314,315],{},"KSampler (Advanced):",[43,317,318,324,330,342],{},[46,319,320,323],{},[14,321,322],{},"Steps:"," 30–50.",[46,325,326,329],{},[14,327,328],{},"CFG:"," 5.0 to 7.0 (Wan2.2 is sensitive; don't go too high).",[46,331,332,221,335,338,339,256],{},[14,333,334],{},"Sampler:",[191,336,337],{},"uni_pc"," or ",[191,340,341],{},"euler",[46,343,344,221,347,256],{},[14,345,346],{},"Scheduler:",[191,348,349],{},"simple",[23,351],{},[26,353,355],{"id":354},"_5-image-to-video-i2v-workflow-tutorial","5. Image-to-Video (I2V) Workflow Tutorial",[10,357,358],{},"The I2V model is the \"gold standard\" for 2026, allowing you to animate static AI-generated art with incredible temporal consistency.",[38,360,362],{"id":361},"step-by-step-implementation","Step-by-Step Implementation",[186,364,365,371,380,390,408],{},[46,366,367,370],{},[14,368,369],{},"Load Image:"," Upload a high-quality source image (e.g., from Midjourney or Flux).",[46,372,373,277,376,379],{},[14,374,375],{},"WanVideo I2V Loader:",[191,377,378],{},"Wan2.2-I2V-14B"," model.",[46,381,382,385,386,389],{},[14,383,384],{},"Image-to-Latent:"," Connect your image to the ",[191,387,388],{},"WanVideo I2V Encoder",". This converts the pixels into the latent space the model understands.",[46,391,392,395,396,399,400],{},[14,393,394],{},"Prompting:"," Describe the ",[14,397,398],{},"action"," only.\n",[43,401,402],{},[46,403,404,407],{},[307,405,406],{},"Example:"," \"The character turns their head and smiles at the camera, wind blowing through hair.\"",[46,409,410,413],{},[14,411,412],{},"Motion Bucket:"," Adjust the \"Motion\" parameter. Higher values (80+) create more aggressive movement; lower values (30-50) are better for subtle portraits.",[23,415],{},[26,417,419],{"id":418},"_6-advanced-optimization-techniques","6. Advanced Optimization Techniques",[38,421,423],{"id":422},"using-lightx2v-v2-lora","Using Lightx2v V2 LoRA",[10,425,426,427,430,431,434],{},"If your generation takes too long, use the ",[14,428,429],{},"Lightx2v V2"," distillation LoRA. This allows you to drop your sampling steps from ",[14,432,433],{},"40 down to 8–12"," without losing significant quality. This is a game-changer for those running on a single RTX 3080 or 4070.",[38,436,438],{"id":437},"vram-management","VRAM Management",[10,440,441],{},"If you run out of memory (OOM error):",[43,443,444,454,461],{},[46,445,446,447,338,450,453],{},"Enable ",[191,448,449],{},"fp8",[191,451,452],{},"bf16"," precision in the loader.",[46,455,456,457,460],{},"Use the ",[14,458,459],{},"\"VAE Tile Encode\""," node to process the video in chunks during the final decoding stage.",[46,462,463,464,467],{},"Reduce the resolution to ",[14,465,466],{},"832x480"," for a faster \"preview\" workflow.",[23,469],{},[38,471,473],{"id":472},"conclusion","Conclusion",[10,475,476],{},"Wan2.2 represents a massive leap in accessibility for high-end video generation. By leveraging the MoE architecture in ComfyUI, you can produce professional-grade clips that rival commercial closed-source tools.",{"title":478,"searchDepth":479,"depth":479,"links":480},"",2,[481,485,486,490,493,496],{"id":28,"depth":479,"text":29,"children":482},[483],{"id":40,"depth":484,"text":41},3,{"id":86,"depth":479,"text":87},{"id":169,"depth":479,"text":170,"children":487},[488,489],{"id":176,"depth":484,"text":177},{"id":209,"depth":484,"text":210},{"id":261,"depth":479,"text":262,"children":491},[492],{"id":268,"depth":484,"text":269},{"id":354,"depth":479,"text":355,"children":494},[495],{"id":361,"depth":484,"text":362},{"id":418,"depth":479,"text":419,"children":497},[498,499,500],{"id":422,"depth":484,"text":423},{"id":437,"depth":484,"text":438},{"id":472,"depth":484,"text":473},"This guide provides a comprehensive breakdown of Wan2.2, the latest evolution in open-source video generation from Alibaba’s Tongyi Lab. Whether you are a professional motion designer or a hobbyist, integrating Wan2.2 into ComfyUI offers unparalleled control over cinematic AI video production.","md",{},true,"/en/advanced-tutorial/wan2.2",{"description":501},"en/advanced-tutorial/wan2.2","ea_HA_HwwZoPLcI9WUQ7SMvTr7mrdDa24HTX6PoKZgo",1773986047107]