A technical web application for exploring the internal data flow and architecture of the WanVideoWrapper and Wan2.1 model(s). Built with vanilla HTML, Tailwind CSS, and JavaScript - no heavy frameworks required.

Features

Pipeline Overview: Interactive visualization of the complete data flow from text input to video output
Text Encoding Analysis: Real-time token counting and T5-XXL embedding visualization
Latent Space Explorer: Understand compression ratios and patchification process
Transformer Architecture: Somewhat superficial but helpful dive into 40 transformer blocks with specialization curves
Memory Analysis: Detailed breakdown of VRAM usage with optimization suggestions
Activation Editor Interface: Visual block selection and blending analysis

Running the App

# From the tools directory
python run_dataflow_webapp.py

# Or directly
python3 run_dataflow_webapp.py

The app will automatically open in your default browser at http://localhost:8080

Technical Accuracy

This webapp is based on the actual implementation details from ComfyUI-WanVideoWrapper:

Requirements

Python 3.x (uses built-in HTTP server)
Modern web browser with JavaScript enabled
No additional Python packages needed

Architecture Details

The webapp accurately represents:

Text encoding through UMT5-XXL
Latent space compression and patchification
Transformer block processing with self-attention and cross-attention
VAE decoding to RGB frames
Memory requirements for different configurations

Configuration Options

Video dimensions (width, height, frames)
Batch size
Model precision (FP32, FP16, FP8)
Attention modes (SDPA, Flash Attention, SageAttn, Radial Sparse)

All calculations update in real-time as you adjust the parameters.