The Best of WebGPU in January 2025

Utsubo, an Osaka-based creative studio, is known for interactive installations and real-time 3D experiences. This site shows how WebGPU can push near-desktop-quality visuals into the browser, from fluid animations to striking lighting effects.

Key Innovations:

-Built with the "WebGPURenderer" from Three.js.
-Features a fully interactive 3D cheetah that moves through various scenes, responding to user interactions.
-Offers a WebGL fallback for older browsers, though Chrome is recommended for the best experience.

Try it here: www.utsubo.com

Imagine instructing a webpage to do all your repetitive tasks with just your voice or typed commands. Jason Mayes has done exactly that with his fully client-side WebAI Agent.

Entirely Local: No server calls—everything runs in your browser, which boosts privacy.

LLM-Powered: Uses Google’s Gemma 2 (2B), a Large Language Model that can understand and generate text using massive training data.

Real-World Demo: Mayes tested it on a flight search site, letting users simply say “Find me a flight to Tokyo next Monday” and have the agent complete it automatically.

Mayes envisions a future where humans and “agents” cooperate on everyday tasks, delivering a more intuitive and private browsing experience.

Watch the demo & code:
- YouTube Video

Suboptimal Engineer merges computational biology with web technology in this captivating slime mold simulation.

Compute Shaders: Small programs that run on the GPU, handling thousands of particle interactions in real time.

TypeScript + WebGPU: Built for the web, so no desktop app needed.

Dev Log & Code: A YouTube video breaks down the technical details for those interested.

This example highlights advanced physics or biology simulations running directly in a browser.

Try the Demo / Code:
- GitHub
- YouTube Dev Log

Matsuoka_601 has developed a fluid simulation based on MLS-MPM, combining physics and cutting-edge browser tech.

MLS-MPM (Moving Least Squares Material Point Method): Allows larger, more stable steps for fluid and soft-body animations.

Particle Counts: Can handle up to 300k particles on mid-range GPUs for real-time performance.

SPH Toggle: Switch between MLS-MPM and Smoothed Particle Hydrodynamics to compare simulation approaches.

This work showcases how computationally heavy physics can still perform impressively in a browser.

Try the Demo: webgpu-ocean.netlify.app

For those wanting to run Large Language Models locally, the new MiniThinky-v2 (1B) demo stands out.

All Local: No server calls; it all runs in your browser with Transformers.js.

Fast: About 60 tokens per second, enough for real-time conversations or logical reasoning tasks.

Extensions Possible: Local AI that reads and manipulates web pages could enable powerful browser-based assistants.

Demo & Source:
- GitHub
- Live Demo

Text-to-Speech transforms written text into spoken audio. This new approach uses WebGPU for faster, more private processing.

500M-Parameter Model: Covers English, Chinese, Korean, and Japanese.

Zero-Shot Voice Cloning: Mimics new voices with minimal input.

Local Processing: No internet is required once it’s loaded, safeguarding your data.

This could help with language learning, improved accessibility, or custom voice assistants.

Demo & Source:
- GitHub
- Live Demo

Axel Johnson is building Realishot, a browser-based renderer aimed at product and packaging visuals.

Ray Tracing: Simulates real-world lighting for more natural shadows and reflections.

Multiple Model Formats: Import existing 3D assets directly.

Figma Integration: Use Figma designs as textures on 3D objects.

AI-Assisted Scenes: Automatically adjust lighting and cameras for easier setup.

Combining Three.js and WebGPU, Realishot delivers near-desktop-level graphics entirely in your browser.

Get the Beta: Realishot.com

Pavel Mazhuga has created a beautiful particle experiment powered by Three.js Shading Language (TSL) and WebGPU.

Compute Shaders, curl noise, and postprocessing (bloom) all come together for a mesmerizing flow field effect.

Live: pavelmazhuga.com/lab/flow-field/webgpu

Source code: GitHub

January 2025 highlighted WebGPU’s versatility, from interactive portfolios and AI agents to fluid simulations and advanced 3D rendering.

The pace of innovation in browser-based graphics and AI keeps accelerating, and we can’t wait to see what else 2025 brings.

- Will we see interactive AR/VR worlds running entirely in-browser?

- Will local AI for text and speech make another leap forward?

Stay tuned as WebGPU continues to reshape real-time graphics and AI on the web. If you have a noteworthy project, send it our way—yours might appear in the next “Best of WebGPU.”