Underdog Demos Woof, a 32MB Speech Model With 31.7M Params Running Offline in a Browser
emmanuelvivier · x · 2026-11-08
While AI companies race to build bigger models, startup Underdog went the other way, releasing a 32MB voice model called Woof and building a browser game around it.
- Scale: Woof has only about 31.7 million parameters, small enough to run in a browser, targeting on-device and offline speech generation.
- Demo: The game is a retro-tech room with a typewriter, a CRT monitor, a desk lamp that clicks on and off, and a music tuner; all voices are generated by Woof.
- Easter egg: The intro warns about "intrusive" cloud providers in town, a jab at cloud-based models; defeating Darkcloud earns an invite.
It represents the opposite path to "bigger is better": tiny models on edge devices.
Related event: Underdog Unveils Woof, a 32MB Voice Model That Runs Offline in the Browser(2 posts)→
More from Multimodal
- Animation built by Claude Opus 5.5 with GPT Image 2.6 textures and code-driven motion — gorkem · 2026-11-09
- PlayCanvas open-sources Gaussian Splat streaming: 121M Gaussians in browser via WebGPU — willeastcott · 2026-11-09
- Halloween short film generated from a single prompt with Seedance 2.5 — LudovicCreator · 2026-11-09
- Full guide: fully local one-shot music videos with ComfyUI Video Builder on 16GB VRAM — Cheap_Credit_3957 · 2026-11-08
- MoCap-to-video pipeline: RGBMoCap + Blender + Seedance 2.5 re-renders K-pop choreography — Choidonhyeon · 2026-11-08
- MiniMax H3 fantasy fight short showcases cinematic video generation — shanytopper · 2026-11-08