openspline/openspline

Give a face to your voice AI in 5 lines of code. Generate realtime avatar video connected to your voice AI locally

What it solves

OpenSpline allows developers to add a synchronized, talking face to existing voice AI agents. It bridges the gap between audio-only AI responses and visual avatars, enabling developers to turn a static portrait and generated audio into a live, moving video stream with minimal setup.

How it works

It uses audio-driven talking-head models to animate a portrait based on incoming audio streams. Developers can integrate it into their existing agent workflows using Python or Node.js SDKs. The system handles the synchronization of audio and video, supports interruption (barge-in) handling, and provides built-in web components for displaying the avatar. It can be run via the OpenSpline Cloud or self-hosted on your own NVIDIA GPU hardware.

Who it’s for

This tool is for developers building conversational AI agents who want to add a visual interface to their voice-based applications. It is designed for those who already have an agent (using providers like OpenAI, LiveKit, Gemini, or ElevenLabs) and want to attach a face to it without rebuilding their existing logic.

Highlights

  • Five-line integration for most voice AI agents.
  • Built-in connectors for popular frameworks like LiveKit, OpenAI Realtime, and ElevenLabs.
  • Supports both cloud-hosted and self-hosted (local GPU) deployment.
  • Includes ready-to-use React and web components for displaying avatars.
  • Handles complex features like audio-video synchronization, idle motion, and barge-in interruptions.

Related

  • Project
    showlab/Kiwi-Edit

    Kiwi‑Edit is an open‑source video‑editing system that takes natural‑language instructions (and optional reference images) to modify whole videos. It fuses a multi‑modal LLM encoder with a video Diffusion Transformer, offering style transfer, object add/replace/remove, and background changes. The repo provides full training scripts (three‑stage curriculum with Qwen2.5‑VL‑3B + Wan2.2‑TI2V‑5B), pretrained Diffusers checkpoints, and evaluation pipelines on OpenVE‑Bench and RefVIE‑Bench.

  • Project
    minnie-lin/Awesome-Physics-Cognition-based-Video-Generation

    This repository organizes research papers, code, and resources on physics-cognition-based video generation. It provides a structured taxonomy, a survey paper, and links to recent work, helping researchers track the field.

  • Project
    zhouwei713/seedance-prompt

    A prompting framework and skill for AI video generators that replaces generic descriptors with detailed simulations of real-world recording equipment and human imperfections to eliminate the "AI look."

  • Project
    microsoft/LatentSpatialMemory

    A framework for video world models that stores 3D scene content as latent tokens to improve generation speed and reduce memory overhead by avoiding repeated RGB rendering.