AutoShorts: turning long recordings into vertical clip candidates with AI moment ranking
AutoShorts is a local-first desktop application for turning long-form video or audio recordings into high-impact, vertical short-form clip candidates (9:16 portrait) with AI-powered viral moment ranking.
At a glance
- What is it?
- AutoShorts is a Tauri 2 desktop app that imports long video or audio, transcribes it, and ranks viral moments into 9:16 portrait clips. The README recommends DeepSeek for detection and warns against small local models for that job.
- Who is it for?
- Adopt AutoShorts if you already cut long recordings into vertical clips by hand and you are willing to hold a Deepgram key for transcription and a DeepSeek key for moment detection. Do not adopt it if you need a fully offline pipeline: the README states local Ollama models are generally not recommended for viral moment detection, and the offline path still requires Python and openai-whisper.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 46 days ago.
- What is it written in?
- Mainly Rust, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 17, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
Who AutoShorts is for, and the manual work it removes
The problem AutoShorts addresses is specific: a long recording holds a handful of moments worth publishing as vertical clips, and finding them by scrubbing a timeline is slow. The repository describes a desktop application that imports media, extracts audio, transcribes it, and then analyzes and ranks moments in what the README calls a single automated chain. Output is aimed at 9:16 portrait H.264 clips rather than a full-length export.
The target user is a solo creator or editor working on one machine, not a team running a hosted service. Projects live in local SQLite, and the README lists transcripts, candidates, custom names and rendering data as what gets saved there. There is a native project manager for creating, opening, renaming and deleting projects from a dashboard. If your workflow already ends in a Premiere or DaVinci timeline and you only need a rough cut, this is a different kind of tool: it produces clip candidates and ranks them, and the README does not describe an editing surface for refining them.
The pipeline: import, transcribe, rank, crop
The stack is Tauri 2 on the desktop shell with React and TSX in the frontend and Rust behind it, with SQLite for storage. The repository layout matches that split: src/ holds the frontend, src-tauri/ holds the Rust side.
Transcription runs through Deepgram in the cloud path. Alternatively the onboarding wizard offers a fully offline option built on Ollama for analysis and openai-whisper for transcription, which the README says requires Python and a pip3 install. Moment detection is handled by a language model, and the README treats the provider as a switch rather than a fixed choice: DeepSeek is the default, Claude is the premium option, and local Ollama models are a third path. The .env.example file lists DEEPGRAM_API_KEY, ANTHROPIC_API_KEY, DEEPSEEK_API_KEY and GROQ_API_KEY alongside LLM_PROVIDER, which defaults to deepseek.
Cropping is delegated to native ffmpeg. The README describes automatic center-cropping of landscape video to vertical portrait clips, and ffmpeg also handles audio extraction and dynamic captions. That dependency is not optional: FFmpeg and FFprobe must be on your PATH before any of the pipeline works.
Installing AutoShorts and running a first project
Users install from the GitHub Releases page, picking the package that matches the machine. On macOS that is aarch64.dmg for Apple Silicon or x64.dmg for Intel; on Windows an .msi or portable .exe; on Linux a .deb or an .AppImage. The README notes the packages are unsigned or self-signed, so both macOS Gatekeeper and Windows SmartScreen will warn. On macOS the documented bypass is a right-click Open, or clearing the quarantine attribute:
xattr -cr /Applications/AutoShorts.appOn Linux, the .deb installs with dpkg and the AppImage needs the executable bit before it will launch:
sudo dpkg -i autoshorts_*.deb
chmod +x autoshorts_*.AppImage
./autoshorts_*.AppImageFFmpeg has to be present first. The README gives per-platform commands, including a Homebrew tap for the build that carries the drawtext and subtitles filters:
brew install ffmpeg
brew tap homebrew-ffmpeg/ffmpeg
brew install homebrew-ffmpeg/ffmpeg/ffmpegOn first launch an onboarding wizard asks you to choose a workflow. The cloud path takes API credentials for Deepgram, DeepSeek and Claude, then a Save & Start button loads the dashboard. The offline path asks you to pick an Ollama model card (llama3.2 3B, qwen2.5 3B or qwen2.5 7B), pulls the weights with a progress bar, and prompts you to run pip3 install openai-whisper. You can change engines later through the API Settings gear icon, or start over with the Reset App Configuration & Onboarding button at the bottom of that panel.
Developers run it from source instead. Copy the environment template, fill in the keys, then start the Tauri dev shell:
cp .env.example .env
npm install
npm run tauri:devA production bundle is built with npm run tauri:build, and the README says the installers land under src-tauri/target/release/bundle/.
Where AutoShorts breaks down
The README is unusually direct about the weakest part of its own design. It states that local models are generally not recommended for viral moment detection, because 3B and 7B models lack the context reasoning needed to evaluate long transcripts and calculate accurate segment timestamps, often returning fragments that are too short. So the offline mode exists, but the project's own documentation steers you away from it for the task the app is named after. That is a real constraint, not a footnote: the fully local configuration is the one most likely to disappoint.
The second constraint is dependency weight. You need FFmpeg and FFprobe on PATH, and the macOS instructions already anticipate a Homebrew build that lacks the filters captions depend on. Caption rendering can fail for reasons that have nothing to do with AutoShorts.
Third, the cloud path sends your audio and transcript to Deepgram and your transcript to a language model provider. A local-first application with local SQLite storage still depends on remote services for the two steps that matter most. If your material cannot leave the machine, the practical pipeline is the one the README discourages.
Finally, the repository carries no licence file among its top-level entries, and no licence is stated. That is worth resolving before you build on it or redistribute a bundle.
How AutoShorts differs from Opus Clip and similar hosted clippers
The obvious comparison is a hosted clipping service such as Opus Clip, which takes an upload, runs detection on the vendor's infrastructure and returns rendered clips through a browser. AutoShorts keeps the project state on your disk in SQLite, runs the crop with your own ffmpeg binary, and lets you point detection at a provider you choose and pay directly. The difference in approach is where the transcript and the render live, and who holds the API relationship.
That trade has a cost. A hosted service hides the FFmpeg install, the PATH problems and the key management behind a subscription. AutoShorts exposes all of them, and the README spends most of its length on exactly those setup steps. It also means cost scales with your own provider usage rather than a flat plan: the README quotes DeepSeek at under $0.001 per transcript and Claude typically at $0.01 to $0.05 per run. Those are the project's stated figures, not measured results.
If you want a tool where the only decision is which file to upload, AutoShorts is the wrong shape. If you want the transcript, the candidates and the renders to stay in a local project you can inspect, the trade is the point.
Version and upgrade cost
The latest release listed is v0.1.5 from 2026-07-01, following v0.1.4 and v0.1.3 within the prior two weeks. The last push to the default branch was on 2026-08-02. The repository is not archived. The version numbers sit below 1.0, which is consistent with a project still filling in its foundation, and package.json shows version 0.1.3, so the manifest and the release tags are not in lockstep.
Upgrade cost is mostly environmental. Desktop installers are replaced by downloading the newer package, and the README does not document a migration path for existing SQLite projects. If the schema changes between releases, the README is silent on what happens to transcripts and candidates you have already generated. Back up the local database before replacing a build, and treat pre-1.0 releases as capable of breaking stored project data.
On licensing: no licence identifier appears in the repository entries listed, and the README does not name one. Absent a licence, the default position is that all rights are reserved, which affects redistribution and any use inside a company. That is a question for your own legal review, not something this article can settle.
Editorial conclusion
Adopt AutoShorts if you already cut long recordings into vertical clips by hand and you are willing to hold a Deepgram key for transcription and a DeepSeek key for moment detection. Do not adopt it if you need a fully offline pipeline: the README states local Ollama models are generally not recommended for viral moment detection, and the offline path still requires Python and openai-whisper. Before installing, verify that ffmpeg and ffprobe resolve on your PATH, and confirm that the Homebrew ffmpeg build on your macOS machine includes the drawtext and subtitles filters, since the README notes the standard formula may lack them.
Frequently asked questions
What is AutoShorts AI?
It is a local-first desktop application that turns long-form video or audio into vertical 9:16 short-form clip candidates, using AI to rank the moments worth clipping. It is built with Tauri 2, React, TypeScript and Rust, and stores projects in local SQLite.
Is AutoShorts AI free?
The application itself is downloaded from GitHub Releases at no stated charge, but the cloud pipeline is not free: you supply your own Deepgram key for transcription and a DeepSeek or Claude key for moment detection and pay those providers directly. The README quotes DeepSeek at under $0.001 per transcript and Claude typically at $0.01 to $0.05 per run.
How do you use AutoShorts AI?
Install FFmpeg and FFprobe on your PATH, download the package for your platform from GitHub Releases, and complete the onboarding wizard. The wizard either takes your Deepgram, DeepSeek and Claude credentials for the cloud path, or walks you through an Ollama model pull plus pip3 install openai-whisper for the offline path, after which you import media and let the pipeline transcribe, rank and crop.
Is AutoShorts AI legit?
It is a real open repository under JayWebtech with tagged releases up to v0.1.5 and a last push on 2026-08-02, so the code is inspectable rather than a closed service. What it is not is finished: the version is below 1.0, the installers are unsigned or self-signed, and the repository lists no licence file.
What are the prices of AutoShorts?
The README states no price for the application itself. The only figures it gives are provider costs for moment detection: under $0.001 per transcript with DeepSeek, and typically $0.01 to $0.05 per run with Claude.
Community notes