Changelog¶

This page tracks OpenTalking releases, capability progress, roadmap items, and compatibility notes.

v0.1.0 - 2026-06-16¶

First GitHub Release Published the initial release package for the OpenTalking orchestration layer, including API, worker, web console, and documentation-backed install paths.
Docker delivery Added versioned GHCR image targets for opentalking-api, opentalking-worker, and opentalking-web.
Python artifacts The release workflow builds and checks wheel and source distribution artifacts for GitHub Release attachment.
Packaging boundary Model weights are not bundled in Python packages or Docker images; users should follow model-specific download and OmniRT/runtime setup docs.

QuickTalk integration QuickTalk / Wav2Lip now have easier startup paths and can be launched directly through OpenTalking for digital-human generation.

MuseTalk WebRTC playback optimization Added MuseTalk media backpressure to improve WebRTC playback stability.

MuseTalk adaptation Added the MuseTalk talking-head path for lightweight full-frame digital-human validation.

Model backend decoupling Decoupled mock, local, direct_ws, and omnirt at the architecture level so different models can choose different deployment backends.

QuickTalk local adapter Added the QuickTalk model adapter, configuration notes, and async initialization.

Baseline real-time digital-human experience Built the main Web console, LLM conversation, TTS, subtitle events, and WebRTC audio/video playback pipeline.

Starting with v0.1.0, this changelog includes formal release sections.
Model integration, runtime backends, and configuration keys are still evolving quickly. Check “Model Support” and “Usage Guide” before upgrading.
Benchmark data must include hardware, model, backend, startup state, and input assets; numbers should not be compared across environments without context.