Wan 2.2 Animate in ComfyUI — Complete Workflow Explained
📂 Workflow Download:
https://drive.google.com/file/d/1S9lj...
⚡ What's covered in this video:
Resolution setup and why global constants matter for clean workflow design
Driving video loading with FPS normalization (force 25 FPS)
Reference image handling — lanczos resize, crop mode, and why previewing your crop matters
Pre-processing pipeline: 3 parallel streams — pose extraction, face extraction, and SAM2 segmentation
YOLOv10M + ViTPose Large Wholebody for full-body keypoint estimation via ONNX runtime
SAM2 video mode for stable, flicker-free segmentation masks across all frames
Grow mask with blur (15px expansion) for hair and clothing edge preservation
Model loading: Wan 2.2 Animate 14B BF16, Sage Attention, block swap (25 blocks) for low-VRAM GPUs
LightX2V LoRA — step distillation that cuts 20–30 steps down to just 4
Wan 2.1 VAE for encoding and decoding (shared architecture with 2.2)
ClipVision embedding of your reference image
Wan Video Animate Embeds — assembling all conditioning signals
Sampling at 4 steps, CFG = 1 (critical for distilled LoRA — do not go higher)
VAE decode with tiling toggle for OOM handling
Motion transfer vs character replacement mode — how to switch between them
📦 Models & Nodes:
🔧 Custom Nodes:
ComfyUI-WanAnimatePreprocess — https://github.com/kijai/ComfyUI-WanA...
🤖 YOLO (Person Detection):
https://huggingface.co/Wan-AI/Wan2.2-...
🦴 ViTPose Large Wholebody:
https://huggingface.co/JunkyByte/easy...
🦴 ViTPose Huge Wholebody (needs both files):
https://huggingface.co/Kijai/vitpose_...
https://huggingface.co/Kijai/vitpose_...
🎬 Wan 2.2 Animate Model (fp8 scaled):
https://huggingface.co/Kijai/WanVideo...
⚡ LightX2V LoRA (Step Distillation):
https://huggingface.co/Kijai/WanVideo...
💡 Wan 2.2 Relight LoRA:
https://huggingface.co/Kijai/WanVideo...
📝 Text Encoder (UMT5-XXL BF16):
https://huggingface.co/Kijai/WanVideo...
🔷 VAE (Wan 2.1 BF16):
https://huggingface.co/Kijai/WanVideo...
📂 Workflow Download:
https://drive.google.com/file/d/1S9lj...
🕐 Timestamps:
0:00 — What is Wan 2.2 Animate?
0:35 — Resolution Group
1:13 — Driving Video Group
1:52 — Reference Image Group
2:35 — Pre-processing: Pose, Face & Segmentation
4:11 — Models Group (Block Swap, LoRA, VAE, ClipVision)
8:03 — Embeddings Generator Group
9:20 — Sampling & VAE Decode
10:09 — Output Group
🔔 Subscribe for more ComfyUI workflows, AI video generation tutorials, and deep dives into AI and software development.
💬 Drop your questions in the comments — I read everything.
#ComfyUI #Wan22Animate #AIVideo #MotionTransfer #CharacterReplacement #WanVideo #AIAnimation #ComfyUIWorkflow #AIContentCreation #StableDiffusionVideo #WanAnimate #VideoAI #OpenSourceAI #ComfyUITutorial #AITools