🎙️ Shipped: TachiDUBB Studio is now open source.
🔥 What's in the box
🎙 Local voice cloning via VoxCPM2 — clones the original speaker, in any of 28 languages
🌍 28 target languages out of the box
🤖 MCP server — Claude Code, Cursor, Cline, any MCP agent can dub videos by tool call
⌨️ Full CLI — cron-able, scriptable, CI/CD-able
🎬 Multilingual showcase reels — N languages → one stitched mp4 with per-language badges
🔧 Per-segment regeneration — fix one bad line in 6 seconds, not the whole video
📦 Batch dubbing — drop a folder, sleep, wake up to dubbed mp4s
🎵 Background music preservation (optional)
🗣 Multi-speaker diarization — clones each speaker independently
🖥 Windows / Linux / headless server
💾 8 GB VRAM minimum
🧠 Persistent job history — re-dub a 6-month-old upload without re-uploading
📜 MIT licensed. Yours forever.
━━━━━━━━━━━━━━━━━━━━
Tell Claude Code:
"dub this YouTube video into French, Spanish and Japanese, stitch a showcase reel."
4 minutes later it says done and tells folder.
No browser.
No API key.
No cloud bill.
100% local. Voice-cloned. MIT-licensed.
👉
github.com/tachikomared/tach…
━━━━━━━━━━━━━━━━━━━━
⌨️ The CLI angle nobody else has
tachidubb showcase ./clip.mp4 \
--langs ru,es,fr,de,ja \
--trim 60 --wait
One command. Five languages. One stitched mp4.
Runs headless on any Linux box.
Schedule it via cron.
Wire it into a GitHub Action.
Try doing that with ElevenLabs.
━━━━━━━━━━━━━━━━━━━━
🤖 The MCP angle the future is built on
claude mcp add tachidubb python tools/tachidubb_mcp.py
Now
@AnthropicAI's Claude Code (or any MCP-aware agent) can dub your videos through natural language.
The repo ships a Claude skill that teaches the agent WHEN and HOW to call the pipeline — comparison vs showcase, when to wait vs poll, how to recover from errors, all of it.
This is what "local agentic" actually looks like.
━━━━━━━━━━━━━━━━━━━━
🔧 The killer feature most tools don't have
Per-segment regeneration.
You finished a 10-minute dub. Segment 47 sounds off. Word translated bad?
Click the segment → regen with a different seed → 6 seconds later it's fixed.
Not "re-render the whole video."
Not "pay for re-dubbing."
Just that one segment.
━━━━━━━━━━━━━━━━━━━━
💸 The math
Built and tuned/tested on a single RTX 3080 Ti (12 GB).
8 GB VRAM minimum. CPU fallback works.
Whisper Ollama VoxCPM2 coexist on one card.
vs Heygen / Rask / ElevenLabs Dubbing at modest creator volume (10-min weekly video × 5 languages):
~$1,800/year saved.
You actually can Dubb 1 hour videos for free.
API keys required: ZERO.
Cloud uploads of your face: ZERO.
Per-minute fees: ZERO.
━━━━━━━━━━━━━━━━━━━━
⚡ Install
Windows: double-click install.bat → double-click start.bat
Linux/macOS: chmod x
install.sh && ./install.sh && ./start.sh
That's it. Browser opens at http://localhost:8910.
Paste a YouTube URL → pick a language → Start.
━━━━━━━━━━━━━━━━━━━━
⭐
github.com/tachikomared/tach…
Star the repo if you'd use it — that's how more people find it.
Built by me and
@smolemaru.
Built with Claude Opus 4.7 mostly.
MIT licensed. Free. Forever.
Support the build, hold
$TACHI on Base 🦀
bankr.bot/discover/0x39b4b87…
@bankrbot @0xDeployer
#BuildInPublic #LocalAI #ClaudeCode #MCP #OpenSource #VoiceCloning #VideoLocalization #CreatorTools #AI #Dubbing #IndieHackers #ossvibejam
Update: TachiDUBB Studio now has an MCP server.
Typed "dub this YouTube video into French and Spanish, stitch a showcase reel" into Claude Code.
It called the tools, watched the jobs, returned the link.
Didn't touch the UI once. Wild.
More in thread 🧵