MiniMax H3 runs as a real-time interactive avatar on Intel Arc Pro B70 GPUs
Opening-Trip9912 · reddit · 2026-09-25
A Reddit user built a pipeline running MiniMax H3 as a real-time interactive avatar instead of offline video generation: user question → LLM answer → speech synthesis → H3 renders the avatar → frames stream back near real-time, with all responses and video generated dynamically and no prerecorded library.
The workload runs locally across Intel Arc Pro B70 32GB GPUs — notable since generative video stacks are almost always NVIDIA-centric. The author is still optimizing latency, GPU scheduling, frame generation, buffering, and audio/video sync.
Next goal: long-running autonomous streams where the avatar talks, answers viewers, and pulls in information without a human behind it.
More from Embodied
- NYU's Steertuning Lets a Robot Learn a Chore From 5 Minutes of Human Demo — CyberRobooo · 2026-11-11
- Boston Dynamics Atlas demoed with Blender 5.3 — TomLikesRobots · 2026-11-11
- Lamport Robotics posts robot 'wake-up test' demo video — R4nsack · 2026-11-11
- Embodied AI Faces Two-Order-of-Magnitude Data Gap as Industry Heads to 10M Hours — GaryMarcus · 2026-11-11
- NVIDIA hosts CoRL 2026 meetup on scaling robot training with Isaac Lab — m_wulfmeier · 2026-11-11
- Lighthouse extends Obsidian welding foundation model to run on any robotic platform — bennash · 2026-11-11