Mosael

Mosael — local-first AI video studio for Mac & Windows

Generate, edit and dub AI video on your own computer.
Bring your own API keys for Seedance, Kling, Veo, Wan and more — or your own ComfyUI — then cut on a timeline, clone a voice, translate and lip-sync, make a photo talk, and turn a product photo into a narrated short.

Download for macOS / Windows · Website · Guide · Workflow templates · WeChat community · Maker on X

English | 简体中文

Mosael is a desktop app that takes a video from idea to upload: AI generation, a multi-track editor, voice cloning and dubbing, digital humans, workflow automation and publishing in one place. It connects to the model providers you already pay for — Mosael doesn't resell credits — and keeps your projects and media on your computer by default.

_{Unaltered captures, composed like the homepage · Original screenshots · Watch the tour · Media credits}

Download

Get the latest stable build from GitHub Releases (users in mainland China can use the Baidu Netdisk mirror). Free for personal, non-commercial use — see License.

Launch the installed app directly. It starts the bundled backend (default 127.0.0.1:8800), loads the frontend and starts the publishing executor; nothing needs to be launched by hand. If a healthy Mosael backend is already listening on port 8800, the desktop app reuses it. Local features work without setup; before using AI chat, image or video generation, voice-over or transcription, add a connection under Settings → AI Chat / AI Image / AI Video / AI Audio, or connect your ComfyUI under Plugins.

What you can make

Official workflow templates are complete pipelines: import one, point it at your own models and footage, and run it. A few of them:

Others cover transcript-based cleanup of talking-head footage, translated dubbing with subtitles only, montages from your own footage, fabric lookbooks, and account and comment analysis.

Features

AI video, image and music generation with your own API keys

AI Studio generates images, video, music and sound effects from text, a first and last frame or reference images. Pick any model from the providers you connect and the controls follow that model; results land in the media library and remember what they were made from. Spending is recorded from the usage and charges providers report. AI Studio · Model connections

ComfyUI client: workflows become models and tools

Connect a ComfyUI server on this computer or your LAN — or let Mosael start one you already installed, or install one for you (Apple silicon, Windows or Linux with NVIDIA). Every saved workflow becomes a model you can pick in AI Studio and on boards, and a tool the agent and workflows can run. The model library lists the server's checkpoints, LoRAs and other files with preview images, base model and trigger words, and downloads what's missing; the workflow library organizes workflows in folders, creates and imports them, and completes missing models and custom nodes. An simplified form shows only the few fields others need to fill in, and the desktop ComfyUI workbench opens ComfyUI's own canvas with Mosael's model library, missing items, form and run results docked beside it. Mosael doesn't replace ComfyUI's own interface — it puts your workflows next to cloud models, the editor and automation. Generating with ComfyUI

AI video editor: timeline, transcript editing and subtitles

Edit on multiple timelines and tracks with overwrite and insert modes, linked picture and sound, ripple edits and the usual J / K / L, I / O and Q / W shortcuts. Cut by deleting words in the transcript, import and export .srt / .vtt captions, stack bilingual subtitles, grade with curves, presets and LUTs, and normalize loudness on export. People and agents can edit the same timeline at once, and ⌘Z undoes only your own step. Editing & color

Voice cloning, AI dubbing and video translation

Dub a subtitle track line by line with a voice cloned locally (F5-TTS or Fish Speech), with a copy of that voice in your own Alibaba Cloud Bailian account for CosyVoice, with free Edge voices, or with OpenAI and Volcengine voices. Split a clip into voice and background, denoise it, or keep only the voice. The translated-dubbing templates transcribe, translate, subtitle and dub a whole video, and can re-sync the speaker's lips. Voice-over

AI digital humans: talking photos and lip-sync

Turn a portrait or a character asset plus a voice-over into a talking video with OmniHuman (Jimeng), Kling Avatar, HeyGen Avatar IV, Hedra Character-3 or Wan, or re-sync the lips of existing footage with Bailian videoretalk, Kling or HeyGen. A real person needs a consent declaration before their face or voice is used. Digital humans

Workflow automation

Connect models, media and tools on a visual node canvas. Required inputs are checked before anything runs, and references that could never resolve are caught up front. Run a flow by hand, on a schedule or from a webhook, or let the agent call it. Workflows · Scheduled tasks

AI infinite canvas

Lay documents, media, character assets, 3D scenes and generations side by side on an infinite board. Connections pass text and reference media to generation; cells transcribe, translate, denoise or split a grid on their own; a timeline cell gives you a rough cut on the board. Comments, mentions and markers keep review anchored. Creative boards

3D previs and Blender

Arrange objects and lighting, keyframe cameras and objects on one timeline, and watch the shot and the camera path side by side. Use the frame, first/last frames or the camera preview video as a reference for the image or video model you choose. Import GLB/glTF and exchange scenes with Blender through its MCP connection; advanced modeling and simulation stay in Blender. 3D scenes & animation

Agent, media library, notes and documents

The agent reads project context and uses tools across media, notes, boards, scenes, editing and workflows; actions that need approval show a confirmation card. Import footage, record your screen or camera, or download a supported video URL; PDFs, Word files, slides and spreadsheets are parsed into text on your machine. Notes keep scripts and research, and a selected passage can be polished, rewritten or translated with each change shown for approval. Recurring characters, locations and props live in the asset library — @ one while generating and its references come along. AI Studio & agents · Media library · Notes · Asset library

Publish to Douyin, TikTok, Bilibili, Xiaohongshu and YouTube

Browser Pool keeps persistent sign-ins and proxies for several accounts. Each platform takes the content it supports — video on every one of them, image-text on Douyin, Xiaohongshu, WeChat Channels, TikTok and X, and articles (columns) on Bilibili. Publishing forms follow each destination's options; review the media, account and copy before submitting, then track the result. Publishing · Browser profiles

Plugins and the Chrome companion

ComfyUI, object storage and MinerU document parsing ship with the app; Blender, Manim, Remotion, TikHub, Baidu Netdisk and more are in the plugin market. Plugins connect local scripts or MCP servers to the agent and workflows. The Chrome video companion reads transcripts in the side panel, translates and imports media. Using plugins · Writing a plugin · Chrome companion

AI-generated content labels

When an export contains AI-generated or AI-altered content, Mosael writes the standard implicit AIGC label into the file and, by default, a visible "AI-generated" mark — in line with China's rules on labelling synthetic content. Turning the visible mark off is your call and is stated in the export dialog.

Supported models and providers

Bring your own API key, or sign in with a supported subscription. Availability depends on your account and region; the full list with model names is in Model connections.

How Mosael fits next to the tools you know

FAQ

Is Mosael free? Downloading it and using it for personal, non-commercial work is free. AI models are billed by the providers you connect, at their own prices. Commercial use requires written permission from the author; the LICENSE is authoritative. For a commercial license, email 1142704468@qq.com.

Is Mosael open source? The source code is public here, so you can read it, learn from it and build it locally, but it is released under a proprietary license, not an open-source one: commercial use and redistribution are not allowed without permission.

Do I need a GPU? Not for Mosael itself. Cloud models run on the provider's servers and ComfyUI runs on whichever machine you connect. Local transcription, voice cloning and voice separation run on your computer after their models are downloaded; keep 10 GB or more of disk space free for them.

Does Mosael upload my videos? Projects and media stay on your computer by default. A cloud model call sends that run's inputs to the provider you chose (some providers only accept public links, so files go to object storage you configure first); URL downloads, online tools and publishing also use the network. With a team server, shared data lives on that server.

Which systems are supported? macOS on Apple silicon and Windows 10/11 x64. There is no Linux installer.

Documentation

Complete user guides live at mosael.com; their source is under website/content/docs/. Implementation references in this repository:

Local development

Requirements

• Node.js 24+
• pnpm
• Python 3.14 and uv
• FFmpeg 8.1+ with ffprobe (required by media import, preview, export, and the complete media test suite)

Install dependencies:

pnpm install
cd backend && uv sync && cd ..
pnpm fetch:tts-python   # the interpreter local engines' environments are built with — same one the app ships

Browser development mode with frontend hot reload:

# Terminal 1: backend (unlocks the data directory's key from the system keychain when needed)
pnpm dev:backend

# Terminal 2: frontend
cd frontend && pnpm dev

Open http://localhost:5173. Electron-only capabilities such as the embedded publishing browser, system tray, and file associations require desktop mode:

pnpm dev

The desktop command starts Vite, the publishing bundle watcher, and Electron together. When what the main process loaded (electron/main.cjs, the preload and the bundles) changes, the running process still has the old code: a notice at the bottom says the main process code was updated, and Restart relaunches only Electron while Vite and the backend keep running (the browser session's top bar shows a restart badge too). The main-window DevTools shortcut on macOS is Cmd+Option+I