Gradio WebGL Viewer Fails to Load Character Model on Windows #14
Description
After a complete and successful installation on a Windows system, the Gradio demo (app_lam_audio2exp.py) launches correctly. The Python backend runs the AI inference successfully when a user provides an image and audio, generating the bsData.json and a final ZIP archive.
However, the frontend WebGL viewer remains blank and does not display the animated avatar. The browser's developer console shows a critical JavaScript error: Error: character model is not found.
This indicates that while the backend is fully functional, there is a bug in the frontend component (gradio_gaussian_render) that prevents it from locating and loading the 3D model assets (e.g., .glb files) on a Windows-based system.
Environment
Operating System: Windows 11
Python Version: 3.10 (via Conda)
PyTorch Version: 2.8.0+cu129 (or similar modern CUDA builds)
Browser: Chrome/Edge (tested via developer console logs)
Steps to Reproduce
Perform a full installation of the project on a Windows machine.
Launch the Gradio demo using python app_lam_audio2exp.py.
Open the provided local URL (http://127.0.0.1:7860) in a web browser.
Open the browser's developer console (F12) to monitor for errors.
In the UI, click on a sample image (e.g., status.png).
Click on a sample audio file (e.g., HillaryClinton_english.wav).
Click the "Generate" button.
Observe the Python terminal, which shows the inference completing successfully.
Observe the browser's developer console, which displays the errors below.
Expected Behavior
The 3D character model should load in the WebGL viewer on the right and begin animating in sync with the audio.
Actual Behavior
The WebGL viewer remains blank. The Python backend completes its task, but the following JavaScript errors appear in the browser console, preventing the rendering from starting.
JavaScript
Error: character model is not found
at as. (index.js:58392:17)
...
Followed by:
JavaScript
Uncaught (in promise) TypeError: Cannot read properties of undefined (reading 'startRenderEvent')
at n (index.js:58685:55)
...
Additional Context
Extensive debugging confirmed that all Python dependencies are correct, the PyTorch installation is compatible with the GPU, and all backend scripts (after being modified for Windows compatibility) are functioning as expected. The issue seems to be isolated to the frontend JavaScript code and its ability to resolve asset paths on Windows.
Thank you for your excellent work on this project!