1. Workflow Overview

This workflow specializes in AI-powered video stylization and generation, featuring:

• Transforming input videos (e.g., dance.mp4) into anime-style sequences

• Frame-to-frame consistency via WAN2.1 model series

• Customizable prompts for character/background control

2. Core Models

3. Key Nodes

• Video Input Group

  • VHS_LoadVideo: Frame loading (supports frame skipping)

  • ImageScale: Resizes to 832x480 (model-compatible resolution)

• Prompt Control

  • Triple CLIPTextEncode: Processes positive/negative/target-style prompts

  • FluxGuidance: Dynamic conditioning strength (value=11 for high stylization)

• Sampling & Generation

  • HYFlowEditSampler: Custom sampler (4th-order DPM++ Solver, seed=123478)

  • VAEDecodeTiled: Tile-based decoding (256x64 tiles)

• Video Export

  • VHS_VideoCombine: Outputs MP4 (16fps, CRF19 compression)

4. Dependencies

• Required Plugins:

  • ComfyUI-VideoHelperSuite

  • ComfyUI-HunyuanLoom (contains custom nodes like HYFlowEditGuiderCFG)

• Model Sources:

  • Download WAN2.1 models from HuggingFace

5. Input Parameters

• Mandatory:

  • Video file path (e.g., dance.mp4)

  • Positive prompt (see example for detailed anime-style description)

• Tunable:

  • CFG Scale (default=7.5)

  • Frame rate (default=16fps)

6. Notes

⚠️ Requires 12GB+ VRAM (video batch processing)
⚠️ Initial run triggers model compilation (~2-5 mins)
⚠️ Outputs saved as hyloom_*.mp4 in ComfyUI/output/