Local MiniMax H3 generates 29 motion clips to build a talking AiTuber avatar
Hailuo_AI · x · 2026-09-08
A Japanese developer built an animated AiTuber avatar using locally-run MiniMax H3, taking a deliberately simple approach: instead of Live2D-style real-time deformation, motion clips are pre-generated in batch and only the mouth is synced to speech at playback.
- Base art from niji·journey; GPT Images produced four-view turnaround sheets, from which H3 generated turns and gestures
- 29 clips total: 16 turns, 4 idle, 9 gestures — batch-generated overnight on an RTX 5090
- Upside: big moves like high kicks, handstands and spins are possible; downside: every angle/expression must be pre-generated
- Better suited to AiTubers where AI picks motions than real-time VTubing
- Stack: Irodori-TTS for voice, gpt-oss-20b for local conversation, AITuber OnAir package for session handling, Codex for implementation; full instructions published on GitHub
More from coding & agent
- ReCache cuts agentic KV memory 92.43% and speeds first token 3.655x with minimal accuracy loss — techNmak · 2026-09-08
- ReCache reuses tool-schema KV states, cutting agent memory 92% with little accuracy loss — techNmak · 2026-09-08
- Weaviate's year of agent memory: why store-everything RAG breaks three ways — victorialslocum · 2026-09-08
- A simple Next.js app built 2 months ago already has 48 vulnerabilities — haydendevs · 2026-09-08
- GPT-6 Astra 'feels like AGI': $1,000 hands-on across real client work, gaming and computer control — PrajwalTomar_ · 2026-09-08
- Google open-sources Artemis, an Android-automation agent scoring 99%+ on AndroidWorld — Shruti_0810 · 2026-09-08