1. Workflow Overview

Designed for Douyin-style smooth outfit transition videos, this workflow uses WAN2.1 video model with GIMM-VFI interpolation to achieve 60fps output. Features image-to-video conversion, dynamic LoRA control, smart block swapping, and dual-format export (MP4/GIF).

2. Core Models

3. Critical Nodes

• WanVideoSLG (Spatial Local Guidance)

  • Param: blocks=9 (processes 9 image blocks separately)

• GIMMVFI_interpolate

  • Key Param: interpolation_factor=3

  • Install: Requires ComfyUI-GIMM-VFI

• WanVideoBlockSwap

  • Swaps 10 attention blocks to maintain background stability

4. Pipeline Structure

graph TB
    A[Input Image] --> B[Feature Extraction]
    B --> C[Block-controlled Generation]
    C --> D[3x Interpolation]
    D --> E[Dual-format Export]

5. I/O Specification

Input Requirements:

• Image Size: 1024×1440 (auto-adjusted to 720P)

• Recommended: Upper-body portraits

Output:

• 30fps MP4 (H.264)

• 16fps GIF (looping)

6. Important Notes

⚠️ Mandatory Setup:

git clone https://github.com/Kijai/ComfyUI-WanVideoWrapper

⚠️ VRAM Requirements:

• Base Generation: ≥8GB

• Interpolation: ≥12GB
  ⚠️ Key Features:

• Uses offload_device for low-VRAM operation

• Preserves facial features during transitions