AutoClip: Free open-source tool turns long podcasts into short clips automatically
The open-source app AutoClip takes a link to a long interview, podcast or talk and turns it into finished clips for TikTok, Reels, YouTube Shorts and more. Subtitles, a thumbnail and a caption are included. Editing and rendering run on your own machine. The AI analysis can use cheap cloud models or run fully locally, for example with Ollama.

What it's about
AutoClip is an open-source app by developer zhouxiaoka that's currently showing up in GitHub's trending list. It tackles a problem anyone who makes video content knows well: digging the best one or two minutes out of a two-hour podcast or interview, reframing it vertically, adding subtitles and tailoring it for each platform takes a lot of time. According to the project description, AutoClip needs nothing more than a link. You paste the URL of a long video, pick your target platform, and the app delivers finished short clips, complete with a thumbnail (cover) and caption text for posting.
Supported platforms include the Chinese services Douyin (China's version of TikTok), Xiaohongshu and Bilibili, plus TikTok, Instagram Reels, YouTube Shorts and regular YouTube. The software is free and released under the permissive MIT license. It's available as a desktop app for Apple Silicon Macs and Windows 10/11 (x64), downloadable at github.com/zhouxiaoka/autoclip/releases/latest. The project page is zhouxiaoka.github.io/autoclip_intro. The app itself costs nothing. If you use cloud AI models with it, the provider bills you based on usage.
What the tool actually does
AutoClip chains several steps into one continuous pipeline. First, it needs a transcript of what's being said. If the original video already has subtitles from its creator, AutoClip uses those, which is the fastest route. Otherwise, it transcribes locally with Whisper or SenseVoice, two speech recognition models that convert speech into text. A cloud service is also available as an alternative. Next, a language model reads the text, picks out the most compelling moments and scores them. This yields up to ten clips per platform that pass the scoring. Additional candidates go onto a backup list, and you can generate them later with a click on "Create this clip".
Every clip gets a cover, title, description, hashtags and a ZIP package containing everything you need to upload. For vertical video, the tool offers two layouts: an interview style (the default for Douyin and Xiaohongshu) and a full-frame podcast style (the default for TikTok, Reels and Shorts). Bilibili and YouTube stay in landscape. The crop automatically follows whoever is speaking. Shots without people are shown in full. By default, AutoClip appends its own branded outro, which you can turn off in the settings. Copied captions, on the other hand, contain no AutoClip mention. If you want to fine-tune things, there's a built-in editor.
According to the developer, it works best with interviews, podcasts, courses and videos where someone talks straight into the camera. For gaming videos or footage with little speech, you can enable image recognition. That requires a model that understands images, and you should still check the selected moments yourself. Sample clips, including conversations with Bill Gates, Jensen Huang, Andrej Karpathy, Sam Altman and MrBeast, are on display in the case gallery. The project says the rights to the originals belong to their creators and the clips are for demonstration only.
Time and cost in the developer's tests
The developer has published his own measurements from October 1, 2026, all on the same Apple Silicon Mac. He explicitly notes that these were three different videos, not a direct speed comparison. The costs cover only Alibaba's qwen-plus text model. Cloud speech recognition, AI image generation and upload services aren't included.
| Version | Source | Result | Text model cost (estimated) |
|---|---|---|---|
| New, with subtitles | Jensen Huang, 1 h 43 min (English → Xiaohongshu) | 7.5 min / 10 clips | approx. 0.09 yuan |
| New, without subtitles | TIM × Luo Yonghao, 2 h 52 min (Chinese → Douyin) | 29.5 min / 10 clips | approx. 0.20 yuan |
| Old | MrBeast, 2 h 06 min (English → TikTok) | 65 min (33 clips) | approx. 0.64 yuan |
In the second test, speech recognition ran locally with the small "Whisper base" model, which explains the longer runtime. The analysis costs amount to pennies. If you use the cloud service on Alibaba's Bailian platform (qwen-audio-3.1-asr-flash) instead of local transcription, the README puts the cost at about 0.1 yuan per hour of audio.
Privacy and model choice
AutoClip stresses that editing and rendering happen on your own computer. What gets sent to cloud services depends on your configuration: cloud analysis sends subtitles and text, image understanding or AI covers send individual video frames, and cloud transcription sends the audio. If you run everything locally, you don't need a cloud API at all. As local options, the project lists Ollama (default address http://localhost:11434/v1) and LM Studio (http://localhost:1234/v1), two apps that serve language models on your own machine. You can also add any OpenAI-compatible service with your own endpoint and key. Finished clips are only uploaded when you publish them yourself. Direct posting works through your own Bilibili account or the Upload-Post service. Anonymous usage stats and crash reports can be switched off.
For testing without an API key, the installer includes a sample project. It also bundles Python and FFmpeg, the well-known video processing tool. If you want local speech recognition, you'll download about 220 MB the first time. On macOS 15 and later, the system blocks the app on first launch. You'll need to allow it under "System Settings → Privacy & Security → Open Anyway". On Windows, SmartScreen may pop up; click "More info → Run anyway".
For tinkerers and AI agents
Besides the desktop app, there's a Docker version with a web interface for your own servers and Linux, plus a command-line interface (CLI) for Python 3.10 or later (3.11 recommended). A typical command looks like autoclip produce talk.mp4 --srt talk.srt --platform douyin --portrait-style podcast. There's also an MCP server. MCP (Model Context Protocol) is a standard that lets AI assistants operate external tools. With it, an agent can have AutoClip generate clips on its own. The new entry points are called start_quick_output and get_quick_output_status. You can install it with uvx autoclip-mcp.
The bigger picture
Automatically cutting long videos into clips is a crowded market. Commercial services, often subscription-based and processing everything in the cloud, have plenty of customers. AutoClip bets on free software, local processing and free choice of models, all the way to fully local operation. That makes it interesting if you'd rather not upload all your footage to someone else's servers, or if you want to save money. Keep in mind, though, that the quoted costs of pennies per video are the developer's own figures and cover only the text model.
A few things deserve a critical look. The project is clearly aimed at Chinese users: the default paths lead to Alibaba services, and the README contains sponsored ads for API resellers with affiliate links. Quality depends heavily on the model you choose, and the developer himself advises checking subtitles, framing and completeness before publishing. Legally important: if you cut up and upload other people's videos, you need the appropriate rights. The tool doesn't take that question off your hands. Also, a single person maintains the project, and there's explicitly no instant support.
Sources
- GitHub Trending github.com
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.