1. Workflow Overview

This workflow generates viral "Cooking Cat" videos for Douyin using Wan2.1 Multiverse's image-to-video capabilities. Key features:

• Dynamic video generation from a single image (e.g., cat cooking)

• LoRA fine-tuning (e.g., "Fat Cat Chef" style)

• 4K upscaling (via 4x-UltraSharp)

• Frame interpolation (10x with RIFE VFI)

• Multi-format output (MP4 + GIF)

2. Core Models

3. Key Nodes

4. Workflow Groups

• Group 1: Image Preprocessing

  • Input: 1152x2048 PNG (cat chef image)

  • Nodes: LoadImage → ImageResizeAdvanced (1024x1024)

• Group 2: Video Generation

  • Core Nodes:

    • WanVideoImageClipEncode

    • WanVideoSampler (DPM++, 15 steps)

    • WanVideoDecode

• Group 3: Post-Processing

  • Upscale: ImageUpscaleWithModel (4x)

  • Interpolation: RIFE VFI (10x)

  • Output: VHS_VideoCombine (MP4 24fps)

5. Inputs & Outputs

• Inputs:

  • Required:

    • Source image (recommended 1152x2048)

    • Prompt (e.g., "A cat eating with a spoon")

  • Optional:

    • Seed (random by default)

    • LoRA strength (default: 0.9)

• Outputs:

  • WanVideo2_1.mp4 (H.264, 24fps)

  • Upscaled GIF animation

6. Notes

• VRAM: ≥12GB recommended (reduced with fp8_e4m3fn quantization)

• Special Features:

  • Fat_Cat_Cook_MZF04 LoRA for viral Douyin style

  • Note node hints: "Test sdpa/flaash optimizations"

• Troubleshooting:

  • Match image ratio to ImageResizeAdvanced settings

  • Verify rife47.pth path if interpolation fails