ComfyUI Wan2.2 Video Loop Generator

A custom ComfyUI node for creating seamless video loops using Wan2.2, with automatic frame chaining between iterations.

Features

• Seamless Video Loops: Automatically chains the last frame of each segment to the next
• Dynamic Prompts: Support for multiple prompts to create varied video segments
• Model Caching: Efficient reuse of loaded models across iterations
• Frame Extraction: Smart last frame detection for smooth transitions
• Progress Tracking: Visual feedback during generation process
• Error Handling: Graceful failure recovery with partial results

Installation

1. Copy this entire folder to your ComfyUI custom_nodes directory:

   ComfyUI/custom_nodes/comfyui-wan-loop/
   

2. Restart ComfyUI

3. The node will appear under video/wan2.2 category as "Wan2.2 Video Loop Generator"

Required Models

This node requires the following Wan2.2 models to be installed in your ComfyUI models directory:

UNet Models (GGUF format)

• Wan2.2-I2V-A14B-HighNoise-Q3_K_S.gguf
• Wan2.2-I2V-A14B-LowNoise-Q3_K_S.gguf

LoRA Models

• Wan2.2-Lightning_I2V-A14B-4steps-lora_HIGH_fp16.safetensors
• Wan2.2-Lightning_I2V-A14B-4steps-lora_LOW_fp16.safetensors

VAE and CLIP

• wan_2.1_vae.safetensors
• umt5_xxl_fp8_e4m3fn_scaled.safetensors

Available Nodes

This package provides multiple nodes for different use cases:

1. Wan2.2 Video Loop (Dynamic Prompts) ⭐ Recommended

• True dynamic prompt inputs - only use the prompt inputs you need
• Individual prompt_1, prompt_2, etc. inputs (up to 10)
• Only connected/filled prompts are processed
• Perfect for variable-length sequences

2. Wan2.2 Video Loop Generator

• Fixed prompt inputs (prompt_1 through prompt_20)
• All optional prompts always visible
• Good for consistent workflows

3. Test Wan Loop Node

• Simple test node to verify installation

Usage

Dynamic Prompts Node (Recommended)

Required:

• start_image: Initial image for the first video segment
• negative_prompt: Shared negative prompt for all segments
• video_length: Number of frames per segment (default: 81)
• width: Video width (default: 832)
• height: Video height (default: 480)
• fps: Output video frame rate (default: 16)

Optional:

• prompt_1: First video segment prompt (required if no other prompts)
• prompt_2 through prompt_10: Additional prompts as needed
• skip_frames: Frames to skip when extracting last frame (default: 80)
• seed: Random seed (-1 for random)

Outputs

• final_video: Complete concatenated video with all segments
• last_frame: Final frame for potential further chaining
• generation_info: JSON string with generation statistics

Example Workflow

Using Dynamic Prompts Node:

1. Add "Wan2.2 Video Loop (Dynamic Prompts)" node to your workflow
2. Load your starting image and connect to start_image
3. Connect individual prompt inputs as needed:
   • prompt_1: "A peaceful lake at sunrise with gentle ripples"
   • prompt_2: "The same lake at noon with bright sunlight"
   • prompt_3: "The lake at sunset with golden reflections"
   • prompt_4: "The lake at night with stars reflecting"
4. Leave unused prompt inputs unconnected
5. Set your video parameters
6. Run the workflow

Automatic Processing:

• Processes only the connected/filled prompts (1-4 in this example)
• Generates 4 video segments seamlessly chained together
• Each segment's last frame becomes the next segment's starting image
• Outputs a single concatenated video loop

Flexible Usage:

• 1 prompt = Single 5-second video
• 3 prompts = 15-second seamless loop
• 10 prompts = 50-second complex transition sequence

Technical Details

Architecture

• Workflow Template Engine: Converts single-shot workflows to reusable templates
• Model Cache: Efficient loading and reuse of Wan2.2 models
• Frame Processor: Handles frame extraction and format conversion
• Iteration Controller: Manages sequential workflow execution

Performance

• Models are loaded once and cached for all iterations
• Each iteration processes independently for memory efficiency
• Progress tracking shows current iteration status
• Supports up to 20 iterations per generation

Error Handling

• Graceful degradation if individual iterations fail
• Partial results returned if some segments generate successfully
• Detailed error messages in generation info output
• Model loading validation with fallback behavior

Troubleshooting

Common Issues

1. "Failed to load models": Ensure all required Wan2.2 models are installed
2. "No video segments generated": Check that prompts are valid and models loaded
3. Memory errors: Reduce video_length or use smaller dimensions
4. Slow generation: This is normal - each iteration takes several minutes

Debug Information

The generation_info output provides detailed information:

• Number of segments generated
• Total frame count
• Model cache statistics
• Prompts used
• Video settings applied

Model Locations

Models should be placed in these ComfyUI directories:

• UNet: models/diffusion_models/
• LoRA: models/loras/
• VAE: models/vae/
• CLIP: models/text_encoders/

Based On

This node builds upon the single-shot Wan2.2 workflow pattern and automates the frame chaining process that would otherwise require manual copy-pasting of nodes.

License

This project follows the same license as ComfyUI.