Show HN: Instant Live Avatars with Lip-Sync on Everyday Hardware (aradzhabov.github.io)

LocalLiveAvatar is a compact real-time avatar system designed to run on everyday hardware—no cloud or high-end GPU required. The author demonstrates instant, lip‑synced video output delivered in chunks for zero perceived latency (with an optional “wait‑for‑complete” mode for offline use). Benchmarks show roughly 1.3 seconds of avatar output per second of CPU time on an AMD Ryzen 9 7845HX and about 5.3 seconds of output per second on an NVIDIA GeForce RTX 5070, highlighting major throughput and energy‑efficiency gains vs. typical avatar stacks. The pipeline supports avatar creation from photos or video, multi‑language text/audio driving, real‑time overlays (background removal, filters, branding) and streaming integration, and can run entirely on local servers for full data privacy. This matters because it lowers the hardware and infrastructure barrier to live, expressive avatars—enabling low‑cost, low‑latency deployments for accessibility tools, live streaming, and potentially robotics or branded avatar marketplaces. The author emphasizes ethical guardrails: no public source release to reduce real‑time deepfake risk, and a preference for vetted commercial partners. The project originated as an assistive service (free for people with disabilities) to create digital twins for those who’ve lost their voice, illustrating both practical impact and why cautious, privacy‑first distribution is central to its roadmap.