1. Workflow Overview

This is an advanced Wan2.1 video generation workflow specializing in "division/replication" visual effects. Key features:

• Dual-modal input (text/image) → video generation

• Object splitting/cloning effects

• Post-processing enhancement

2. Core Models

3. Key Components

3.1 Custom Nodes

1. ComfyUI-WanVideoWrapper (Required):
   - Contains all Wan video processing nodes
   - Requires manual install: `sageattn` optimizer

2. Custom LoRA:
   - `Wan2.1_division_replication.safetensors`
   - Recommended strength: 0.9-1.1

3.2 Critical Nodes

• WanVideoBlockSwap (ID:32): Controls frame block swapping

• WanVideoTeaCache (ID:98): VRAM optimization

• WanVideoSLG (ID:96): Smooth Latent Guidance

4. Workflow Stages

Phase 1: Input Processing

• Text: "A smartphone splits into multiple apples"

• Image: 480x720 resolution (auto-adjusted)

Phase 2: Feature Encoding

• Text → umt5 encoder

• Image → CLIP vision encoder

Phase 3: Video Generation

• UniPC sampler (30 steps)

• SLG smoothing (strength 0.6)

Phase 4: Output

• 272 frames @16fps

• Automatic VRAM cleanup

5. I/O Specifications

Input Parameters:

{
  "positive_prompt": "object splitting effect", # Required  
  "negative_prompt": "low quality...",  
  "seed": 594591065694543,          # Fixed seed  
  "resolution": [480,720]           # 16:9 recommended  
}

Output:

• MP4 video (H.264)

• Metadata with generation params

6. Important Notes

1. Hardware:

   • Minimum: RTX 3090(24GB)

   • Recommended: A100 40GB

2. Optimization:

   • Enable offload_device

   • Use fp8_e4m3fn quantization

3. Troubleshooting:

   • VRAM overflow → Reduce block_swap frequency

   • Frame flickering → Increase SLG strength