OpenSpline: free local talking-avatar software; bring your own NVIDIA GPU

The offer

Developers in India can download OpenSpline free to turn a portrait and their voice-agent audio into a talking avatar on their own NVIDIA GPU, using Linux or WSL2. The application and current Python SDK 0.1.1 are MIT-licensed, but bundled inference components and downloaded models have separate licences, including Open-RAIL-M for the Lite model VAE. This is self-hosted software, not free cloud avatar minutes: you supply compatible hardware and pay any separate voice, language-model or cloud-provider bills; AMD and Apple GPUs are not supported.

Who can claim it

• Self-host on Linux x86_64 or WSL2 with a compatible NVIDIA GPU and driver.
• The Python SDK supports Python 3.10–3.12; the Node SDK requires Node 20 or newer.
• Open-source software has no India restriction; model and connected-service terms apply separately.

How to claim it

1. Open the official repository and read the Self Host and GPU compatibility sections.
2. Review the installer and install the GPU service on your own Linux or WSL2 machine.
3. Start the service and open localhost:7860 for the demo.
4. Choose a portrait you have permission to use and connect your existing voice agent or audio stream.

Good to know

There is no published software trial expiry. Self-hosting continues to require your own hardware and electricity; hosted OpenSpline Cloud and any connected AI or voice providers are separate services.

MIT covers OpenSpline, not every dependency or model. Vendored inference code is Apache-2.0. SoulX-FlashHead transformer and Pro VAE weights are Apache-2.0; the Lite VAE requires Open-RAIL-M compliance. Performance depends on hardware, and Blackwell/pre-Ampere support still needs validation. Do not infer free cloud minutes or bundled paid model access.