BaoCut

BaoCut is a desktop video editor from Jim Liu. You tell it what you want in plain words, and it transcribes, subtitles, translates, dubs, trims filler words and pauses from talking-head footage, and can build explainers or product videos from a prompt. The README puts the problem this way: most AI video tools render once and hand you a file, so a wrong second sentence in the translation means starting over. BaoCut treats the video itself as the source of truth, with subtitles, voice-overs and motion graphics on one timeline; fix one subtitle line and only that line changes. You and the AI share the same editor, so your drag and its caption edit go through the same undo stack, and you can take over at any point. Transcription, speech and images can run on local models or on online providers such as OpenAI, Anthropic, Google, ElevenLabs, Qwen and Volcengine, and a ledger tracks what each job costs. Beyond the desktop app there's a CLI, a web client, and an MCP service so your own coding agent can drive it.

Who it's for:creators who make talking-head, explainer or bilingual-subtitle videos and want AI to handle the grunt work while they keep the final say.

Before you start: most people should just download the installer, for macOS 14+ on Apple Silicon or Windows x64 (CPU, CUDA and Vulkan builds; the Windows packages are unsigned and need the Microsoft VC++ redistributable first). You also need ffmpeg and ffprobe installed separately for analysis, transcription prep and export, and the chat-to-edit features need a signed-in Codex CLI or Claude Code on the same machine. Local models download from Hugging Face or ModelScope on first use; if you pick an online provider, your text and media go to that provider, which bills you. We don't recommend the run-from-source route for non-coders: it pipes a remote install script straight into the shell with curl and may use sudo to change system settings. The CLI commands baocut skill install and baocut mcp install change your AI agent's configuration, so know that before you run them. Get the owner's consent before cloning anyone's voice; some local models (such as OmniVoice, Fish Audio S2 Pro and Qwen-Image) are non-commercial only, and the app flags this before you pick them. On licensing, personal use, study and internal company use are free, and so is making ads, paid videos or client work with BaoCut; selling or rebranding the software, offering it as a hosted service or embedding it in a commercial product needs a separate written license.

Based on the project's public documentation. Not tested by AI Inspiration World.