1. Workflow Overview

This is an Image-to-Video (I2V) workflow based on WanVideo, designed to generate videos simulating a "sticker peeling" effect. Key features:

• Image understanding (via Janus model for auto-prompting)

• Video frame generation & interpolation

• Optional upscaling (4x resolution enhancement)

2. Core Models

3. Key Nodes & Installation

Plugin Install Command:

# Via ComfyUI Manager
"ComfyUI-WanVideoWrapper", "janus-pro", "comfyui-videohelpersuite"

4. Workflow Structure

Group 1: Image Input & Preprocessing

• Input: User-uploaded image (e.g., output (7).png)

• Process: Resize to 1536px (long edge) while maintaining aspect ratio

• Output: Normalized image

Group 2: Prompt Generation

• Auto Mode: JanusImageUnderstanding generates prompts (e.g., "A girl on the street being peeled off")

• Manual Mode: Direct input via StringConstantMultiline

Group 3: Video Generation

• Core Node: WanVideoSampler (25 steps, UniPC scheduler)

• Effect Control: WanVideoLoraSelect loads LoRA (wan-绮梦流光-解压神器之撕掉贴纸人)

Group 4: Post-Processing

• Upscaling: Optional 4x super-resolution

• Output: MP4 file (16FPS, H.264)

5. Input & Output

• Inputs:

  • Image file (PNG/JPG)

  • Optional manual prompts

  • Resolution limit (default: 1536px)

  • Seed (fixed or randomized)

• Output:

  • MP4 video (e.g., AnimateDiff_xxxx.mp4)

6. Notes

1. VRAM: Minimum 12GB (16GB+ recommended)

2. LoRA Toggle: Manually enable lora1/lora2/lora3 nodes (Right-click → Enable)

3. Troubleshooting:

   • Blurry characters: Increase denoise (default: 0.8)

   • Lag: Reduce batch size (default: 3 → 1)

4. Compatibility: Update ComfyUI-WanVideoWrapper to the latest version.