Violin¶

Development line: project:violin · thread violin
Last event: 2026-05-15 · 1 dated since 2026-05-15 · Researched: 2026-09-05 · confidence: medium

What it is¶

Violin is an MIT-licensed CLI, FastAPI app, and Claude Code skill for video translation and dubbing.

• Transcription: extracts speech from video.
• Translation: translates dialogue across target languages.
• Dubbing: generates replacement voice audio.
• SRT subtitles: creates timed caption files.
• Video Q&A: answers questions about video content.

Supports 33 target languages; six style profiles marked experimental.

Practical for permitted video, though the public demo warns of possible translation errors.

Development line¶

• 2026-05-15 — Public introduction of the Violin translation project. On 2026-05-15, Violin was publicly associated with a Together AI article, an open GitHub repository, and its project website. The dated links indicate a public introduction of an open-source, translation-oriented project and establish its initial public references.

What changed¶

2026-05-15 — Violin appeared as an open pipeline combining ASR, LLM translation, and TTS. The primary source dates the post to 2026-05-14 instead of 2026-05-15. It listed the original models as Together Whisper Large V3, DeepSeek V4 Pro, and Cartesia Sonic 3. The current repository shows version 0.1.1, 33 languages, swappable Together/OpenAI/ElevenLabs providers, and no GitHub Releases. We cannot reliably date those later changes from available primary sources.

How to use this¶

From 2026-05-15, practitioners should consider Violin a publicly reachable translation-oriented project and assess it through its linked repository and project site.

1. For local use, install Python 3.10+, ffmpeg, and the package via uv tool install violin. Set a provider key, then run violin input.mp4 output.mp4 --language Chinese. — https://github.com/shang-zhu/violin
2. For the web UI and REST API, run violin-api. Create a job via POST /jobs, poll it, and download the video or SRT file. — https://github.com/shang-zhu/violin
3. For a quick check, use the web demo. Upload a file or paste a URL, then pick a language, voice, subtitles, and voice-over mode. — https://www.violin-ai.com/

Best practices¶

• Translate only owned, public-domain, or Creative Commons media, or content with explicit permission. The demo restricts input URLs to this rule. — https://www.violin-ai.com/
• Start with the default profile and automatic voice selection. Use a separate YAML config for terminology and check the output, because the demo is marked as a research beta. — https://github.com/shang-zhu/violin
• In production, use config/prod.yaml. It limits uploads, serializes jobs, and caps ffmpeg concurrency. — https://github.com/shang-zhu/violin

Superseded by this¶

• 2026-05-14 — The initial description defaulted to DeepSeek V4 Pro and Qwen3.5-397B-A17B. The current README specifies DeepSeek-V4-Pro-0813 for translation and GLM-5.3-Flash for chat, but no date confirms the switch.
• 2026-05-14 — Missing voice cloning remains an active limit. The repository still lists it under TODO rather than as an available feature.

Still unknown¶

• Available primary sources provide no dated history after launch. We do not record current model settings as new dated events.
• GitHub shows no releases. Version 0.1.1 appears in pyproject.toml, but available sources confirm neither a release date nor a PyPI distribution.
• Users must evaluate dubbing quality on their own language and footage. The public demo warns that translations may contain errors.

Sources¶

Agent brief¶

• Subject: project:violin, thread violin, 1 dated events 2026-05-15 → 2026-05-15.
• Practical note: From 2026-05-15, practitioners should consider Violin a publicly reachable translation-oriented project and assess it through its linked repository and project site.
• Confidence: medium. Dated supersedes above are the authority for what is obsolete.