Media Processing

Immich118.4k★

Immich is a self-hosted Google Photos replacement: automatic backup from your phone, a fast timeline, shared albums, a map view, and search by faces, places, objects or plain-language description, with the machine learning running on your own server. AGPL-3.0, and stable since its 2.0 release in October 2025. Docker Compose is the recommended install. Plan for at least 6 GB of RAM (8 recommended) and 2 cores, keep the Postgres database on a local SSD rather than a network share, and budget 10 to 20% extra disk for thumbnails and transcoded video on top of the library itself. Since v3, the machine-learning container needs a newer x86 CPU (x86-64-v2). Every feature is free, and the team has promised nothing will ever go behind a paywall. You can buy a lifetime supporter key ($25 for one user, $100 for a family), and it unlocks no extra features. Solo users and families get the most here. It has multi-user and OAuth login, but it's built for households, not company-wide photo management. The catch: you're now the backup. Follow 3-2-1 (three copies, two kinds of media, one offsite), because a self-hosted photo library with one copy is one dead drive from gone. Major versions also break third-party API integrations, so check your add-ons before upgrading.

whisper112.1k★

Whisper is OpenAI's speech-to-text model, and it's still the engine most open source transcription tools are built on. Feed it audio in any of about 100 languages and you get a transcript, or an English translation. Code and weights are MIT, so running it on your own machine costs nothing. Setup is pip install -U openai-whisper plus ffmpeg. Hardware is the real cost: tiny and base need about 1GB of VRAM, the default turbo model about 6GB, large about 10GB. Turbo is fast but can't translate. CPU works, slowly. Most people now run the same weights through faster-whisper (up to 4x faster) or whisper.cpp on a Mac. OpenAI's hosted Whisper API is $0.006 a minute, and its newer gpt-4o-mini-transcribe is half that at $0.003. Solo: hosted for the occasional file, local when the audio is private. Small teams: local on one GPU. Larger teams with real volume: self-host and the per-minute bill disappears. The catch: this repo gets small fixes but no release since June 2025, it does no speaker labels or live captions, and OpenAI's own model card warns it can hallucinate words nobody said. For a desktop app, use Buzz or Vibe. For speaker labels, WhisperX.

VoiceStudio102.3k★

VoiceStudio (formerly OmniVoice Studio) is a local, open source answer to ElevenLabs. It clones and designs voices, does text-to-speech across 16 engines and transcription across 11, dubs video by transcribing, translating and re-voicing it, builds audiobooks, and runs a system-wide dictation widget. The app is AGPL-3.0, free, and runs on your own machine. A GPU is optional but makes the difference. Minimums are 8 GB of RAM and 10 GB of disk, and the default workflow wants 8 GB or more of VRAM on NVIDIA or Apple Silicon. ROCm is Linux-only and opt-in. Installers cover Apple Silicon Macs, Windows 10/11 and recent Linux; Docker images are amd64 only, and Intel Macs need a remote backend. Solo creators and small teams doing voiceover, dubbing or audiobooks get no per-character billing and no audio leaving the machine. It also exposes a local REST API, an OpenAI-compatible audio API and an MCP server. Companies embedding it in a closed product can get a commercial license by enquiry, with no published price, and a hosted Cloud API is in early access. The catch is the model licenses, not the app's. The default voice engine's pretrained weights are CC-BY-NC, noncommercial, so paid voiceover work means switching to an Apache-2.0 engine like CosyVoice 3 or VoxCPM2. It is still labeled active beta. Watermarking is on by default, but responsible voice cloning is on you.

OpenCut97.4k★

OpenCut is a free video editor that runs in a browser tab: a multi-track timeline, text, stickers, masks, keyframe animation, effects and auto captions, then export. MIT licensed, no paid tier, no account. It exists because CapCut paywalled basic editing, and its privacy policy says your media never leaves your device, captions included. Most people just open opencut.app and start cutting; projects live in your browser's storage. Running your own copy means the classic codebase: a Next.js app with Docker Compose for Postgres and Redis, plus a Rust core compiled to WebAssembly. One docker compose up -d does it. Solo creators and small teams doing straightforward cuts and captions get what they need for free. Teams that want transcript-driven editing or shared review should pay for Descript or CapCut Pro. There is no tier to grow into. The catch: the live site still runs the classic build (v0.3.0, April 2026), which its own repo marks archived and no longer maintained. The from-scratch rewrite, promising desktop, mobile, plugins and an MCP server, has barely moved since early August, and its desktop app is, in the project's words, just a window that opens. Nothing you can use today is getting fixed.

hyperframes79.8k★

HyperFrames turns HTML, CSS and animation libraries like GSAP, Lottie and Three.js into MP4 videos, frame by frame in headless Chrome, so the same input always renders the same video. HeyGen built it so AI coding agents can write videos as plain web pages, and uses it in production. It's Apache 2.0 with no per-render fees or headcount limits. Run npx hyperframes init, preview, render. You need Node 22+ and FFmpeg. Rendering locally is free; scaling out means building your own AWS Lambda or Cloud Run setup with their packages, or paying HeyGen for managed cloud rendering by the credit. It ships agent skills for Claude Code, Codex, Cursor and Gemini CLI, including one that ports Remotion compositions over. The obvious comparison is Remotion (remotion-dev/remotion), which is source-available and free only for individuals and companies of up to 3 employees; beyond that it's $25 a seat a month for creators or a $100 monthly minimum for automators. Any team of four or more building programmatic video should look hard at HyperFrames for that alone. Remotion still has the more mature cloud renderer. The catch: it's an on-ramp to HeyGen. The default media skill wants you signed into the HeyGen CLI for music, images, voice and avatars, cloud rendering bills HeyGen credits, and the CLI sends usage telemetry unless you turn it off. It's pre-1.0 and ships several releases a day, so pin your version. And hyperframes.app is a lookalike site, not HeyGen.

OBS Studio79.0k★

OBS Studio is a free capture-and-broadcast desk for livestreams, screen recordings and tutorials. Scenes, sources, live compositing, multi-track audio, and output to a local file or a stream over RTMP, SRT or WebRTC. GPL-2.0-or-later, funded by donations and sponsors, with nothing to upgrade to. Download and run on Windows 10 or 11, macOS 13 or newer, or Linux via Flatpak or the Ubuntu PPA. Hardware encoding covers NVIDIA, AMD, Intel, Apple's VideoToolbox and VA-API on Linux, and a virtual camera plus a WebSocket remote-control API ship built in. The real cost is the first hour: OBS hands you a mixing desk, not a record button. Solo, small team or large: same answer. Free, no seats, nothing to administer. Want a share link and a two-click workflow? Cap fits better. A polished product demo with automatic zooms? Recordly. The catch: OBS doesn't edit and has no sharing layer, which is what Loom charges $18 per user per month for. Treat browser-source overlays as untrusted too. In September 2026 researchers ran code on a streamer's PC from a Twitch chat message via a buggy overlay, because OBS 32.2.2 bundles a 2024 Chromium with its sandbox disabled. The 33.0 release candidate updates it.

Remotion71.1k★

Remotion lets you make videos by writing React components. Each frame is a render, and Remotion turns the sequence into an MP4. That makes data-driven and personalized video practical: connect a spreadsheet or an API and render thousands of variations instead of editing each one by hand. One command, npx create-video@latest, gets a project running. Rendering happens where you choose: Node.js on your own servers, in the browser, or serverless on AWS Lambda, where the compute lands on your cloud bill. The documentation runs past a thousand pages, with templates, captions, transitions and agent skills. Company size decides who pays, not features. Individuals, nonprofits and companies of up to three people use it free, even commercially. Bigger companies need a license: $25/mo per seat for making videos yourself, or $0.01 per render with a $100/mo minimum for apps that generate video automatically. The paid version has the same features as the free one. The catch is the label. Remotion is source-available, not open source, and says so itself. You can read the code but not resell a derivative, a fourth person on the team flips you from free to paid, and codec patent licensing for formats like H.264 stays your responsibility. Just trimming clips or burning in captions? FFmpeg is simpler.

FFmpeg66.6k★

FFmpeg is the engine under most of the audio and video software you've touched. Convert formats, cut clips, resize, pull audio, stream: it does it all from the command line, and its libraries sit inside VLC, OBS and HandBrake. Free, no company behind it, no paid tier. Nothing to host. apt install ffmpeg or brew install ffmpeg; on Windows the project points you to builds from gyan.dev and BtbN. Current stable is 9.0, and recent releases added GPU codecs that run on Vulkan and a Whisper filter for speech-to-text. Solo and small teams: free, wrap common jobs in scripts. Larger teams: free, the cost is compute and whoever owns the pipeline. Licensing is where people get burned. The default build is LGPL 2.1+, fine inside closed-source apps if you link dynamically and ship the FFmpeg source. Enable x264 or x265 (the popular H.264 and H.265 encoders) and the whole build becomes GPL, as Homebrew's does. Enable nonfree parts like Fraunhofer's AAC encoder and you can't redistribute it at all. No commercial license exists, at any price. The catch: the command line is brutal. Basic conversion is one line; filters, stream mapping or hardware encoding mean docs and trial and error.

PhotoPrism40.4k★

PhotoPrism gives you automatic face recognition, location mapping, and AI-powered search across your photo library without sending your photos to anyone's cloud. Point it at a folder of images, and it indexes everything with object detection, color analysis, and GPS data. This is the most popular self-hosted photo management tool. It handles RAW files, videos, live photos, and generates thumbnails automatically. The search is good: type "beach" and it finds your beach photos without you tagging anything. The Community Edition is free. PhotoPrism Essentials ($3.99/mo) and PhotoPrism Plus ($7.99/mo) add features like enhanced video support, premium themes, and priority access to new features. The AI features (face recognition, search) are in the free tier. The catch: self-hosting PhotoPrism is not trivial. It needs Docker, a database (MariaDB recommended), and decent hardware for the AI indexing; a Raspberry Pi 4 works but initial indexing of a large library will take days. RAM usage is real: 4GB minimum, 8GB+ recommended for large libraries. And the license is AGPL-adjacent (custom); check terms carefully if you're building a product on top of it.

ente29.9k★

Ente Photos replaces Google Photos and iCloud Photos, except your pictures are encrypted on your device before upload, so Ente can't see them. The same codebase ships Ente Auth, a free 2FA app that fills the hole Authy left. Hosted, you get 10GB free forever, and the whole stack, server included, is AGPLv3. Self-hosting runs the same code Ente runs in production, no stripped edition. A quickstart script brings it up with Docker Compose on a 1 GB RAM box. The chores: set up SMTP for signup codes, lift storage caps with the Ente CLI, and own your backups. Paid plans are storage tiers, from 50GB at $2.49/mo to 2TB at $19.99/mo, and one plan covers you plus 5 family members. Solo: the free 10GB. Families: 200GB at $4.99/mo split six ways is hard to beat. Small teams and up: this is a personal photo service, not a company drive. The catch: Ente itself can't rescue you. Lose your password and recovery key and only a trusted contact you added in advance through Legacy can get you back in, after a 7 to 30 day wait. Let a paid plan lapse and everything is deleted after a 30-day grace period.

faster-whisper26.5k★

faster-whisper runs OpenAI's Whisper speech-to-text models up to four times faster than the original implementation while using less memory, with the same accuracy. It reimplements inference on CTranslate2, and it has quietly become the engine inside most self-hosted transcription stacks. MIT licensed, completely free. It's a Python library, not a service: pip install, pick a model size, feed it audio. A GPU gets you faster-than-realtime transcription; CPU works fine with the small and medium models. Wrapping it in an internal API for your team is an afternoon with FastAPI, and 8GB of VRAM comfortably runs the large model in int8. Anyone paying AssemblyAI or Deepgram per minute for plain transcription should run the math. A cheap GPU instance chews through hours of audio for pennies, and openai/whisper accuracy is the same thing you're renting. The paid APIs keep winning on streaming, speaker diarization, and zero-ops. The catch: the repo has gone quiet. The last release was v1.2.1 in late 2025 and there are hundreds of open issues. It still works, and CTranslate2 underneath it is stable, but nobody is shipping fixes right now. You also own the pipeline: chunking long files, retries, scaling workers, word-level timestamps. The per-minute APIs charge for exactly that boredom, and for being maintained.

Cap24.5k★

Cap records your screen and camera and hands you a share link, which is Loom's whole business. The desktop app runs on macOS, Windows (beta) and Linux (.deb, AppImage or RPM), the platform is AGPLv3, and you can point the app at your own server or S3 bucket instead of Cap's cloud. Personal use is free. Self-hosting starts with one docker compose up -d: Cap Web, a media server doing the FFmpeg work, MySQL and MinIO. Production means real secrets, SSL, Resend for login emails, and your own API keys for transcripts. Cap offers no direct support for self-hosted deployments. The paywall is a license. Any commercial use of Cap's prebuilt app needs the Desktop License at $29/year or $58 once. Cap Pro is $12 per user per month, or $8.16 billed annually, and adds unlimited share length, AI transcripts and the Loom importer. SAML SSO is a flat $199/mo add-on. Solo: $29/year once it's work. Small teams: Cap Pro. Larger teams: Pro plus SSO, or self-host. The catch: free and Desktop License share links stop at five minutes, too short for most demos. Building the app from source skips the license, but then you own the build.

buzz22.4k★

Buzz is the desktop app that makes Whisper usable for people who do not live in a terminal. It transcribes and translates audio and video files, YouTube links, even live microphone input, all offline on your own machine. It is MIT licensed and free, and it runs on macOS, Windows, and Linux. Under the hood it is OpenAI's Whisper with a real UI bolted on: speaker separation for messy audio, export to SRT, VTT, and plain text, a watch folder that auto-transcribes new files, and GPU acceleration whether you are on an Nvidia card, Apple Silicon, or Vulkan. You download it and go. No Python environment, no ffmpeg wrangling. There is a paid Buzz Captions build on the Mac App Store for a few dollars if you want a native, auto-updating package, but the open-source app does everything and costs nothing. Solo users and small teams: grab the free build. This is the easy on-ramp to local transcription, and paying for the convenience version is fair if you want it. The catch is that quality and speed are still Whisper's. A big model on weak hardware is slow, and a small model on hard audio makes mistakes. Buzz makes Whisper convenient; it does not make it faster than your GPU allows.

Kap19.4k★

Kap was the friendly macOS screen recorder: open source, small, drag a frame around the region you want and export a GIF or an MP4 in a couple of clicks. Built with web technology, MIT licensed, and a pleasure to use when it was current. That last clause is doing a lot of work. The most recent release shipped in October 2022, and one of its notes was fixing a crash on the macOS 13 beta. The last commit of any kind landed in late 2024. For an Electron app on macOS, a platform that reworks screen capture permissions and APIs on a regular schedule, that is a long stretch with nobody at the wheel. It still installs and still works for plenty of people, so this is not a claim that it is broken. It is a claim that nobody is on the other end when it does break. Anyone already using it happily: carry on. Anyone choosing a recorder today: CapSoftware/Cap for share links, webadderallorg/Recordly for polished product demos, or obsproject/obs-studio if you want the one that will still be maintained in five years. The catch is the whole review. Kap is good software that stopped getting attention, and open source hands you the source code, not somebody's time. Anything in your daily workflow needs a maintainer behind it, and right now this does not have one.

subtitleedit14.9k★

SubtitleEdit is the desktop tool video editors and translators have used to fix subtitles for over fifteen years. You load a video, see the audio waveform, and edit subtitle timing and text against it. Handles 300+ subtitle formats, batch conversions, OCR for image-based subs, and optional integrations with translation and speech-to-text APIs. MIT licensed, fully free. It is a desktop app. Windows users get a clean installer. macOS users have to install mpv and ffmpeg through Homebrew, then clear the unsigned-app quarantine flag the first time they launch it. Linux users can use Flatpak or install through their package manager. For one-off subtitle fixes, batch format conversions, or cleaning up raw transcription output, this is the standard. Paid cloud tools like Kapwing or Rev offer one-click translation and automated transcription, but you pay per minute and lose the format flexibility. The catch: the UI looks like Windows software from 2010 because that is the era it is from. Functional, dense, no design system. If you want something that feels modern, this is not it.

palmier-pro14.7k★

Palmier Pro is a native macOS video editor with AI wired directly into the timeline. The editor itself is open source under GPLv3 and free to use with no login. Where it gets interesting: you can generate video and images from models like Kling and Seedance right inside your edit, and it hooks into AI agents (Claude, Codex, Cursor) over MCP so an assistant can edit the project alongside you. There's not much to "host" here, it's a desktop app you install and run. The setup catch is hardware: it's Apple Silicon only and requires macOS 26 (Tahoe), so it won't run on older Macs or anything non-Apple. The core editing, the MCP server, and the agent chat are all in the open source build. The free editor is a capable tool on its own. The generative AI features are the paid part: those run closed source and need a login and a subscription. So the split is clean, free editing forever, pay only when you want in-timeline generation. Solo creators and developers curious about agent-driven editing: the free build is worth a look. If you'd lean on the AI generation daily, price the subscription against how much you'll actually use it. The catch is that the headline feature, the AI generation, is exactly the piece that's closed and paid. The open source part is a solid editor; the "built for AI" magic is a subscription. That's a fair way to monetize, just know going in that "open source AI video editor" is half the story.

Concat14.5k★

Concat is a desktop video editor built to do what people actually use CapCut for, without the watermark or the subscription. Multi-track timeline, split, trim, merge, transitions, speed control, titles, and reusable templates, with a Rust engine underneath and a React interface on top. Auto-captions and text-to-speech run entirely on your machine, so your audio never leaves it. Mozilla Public License 2.0, installs and runs with no account. Nothing to host. Download a build for macOS, Windows or Linux and start cutting. The macOS binaries are unsigned, so first launch needs xattr -dr com.apple.quarantine /Applications/Concat.app, which is a small tax for not paying anyone. Local captioning and TTS means CPU work happens on your hardware rather than someone's GPU cluster, and that is the whole point. Free with no tiers. Creators cutting short-form video get the obvious win: no watermark, no export limit, no account. Teams doing production work should still reach for a mature NLE. Compare against OpenCut (OpenCut-app/OpenCut) if you want another young open source editor, or keep Auto-Editor (WyattBlue/auto-editor) around for the scripted cuts. The catch is that this is alpha software at version 0.2.0 from a very new project. It will crash, formats will surprise it, and a timeline you spent three hours on is not something to trust to a pre-release editor yet. Export early and keep the source files.

openwhispr10.8k★

OpenWhispr is voice-to-text for your desktop. Hit a hotkey, talk, and it types into whatever app you're in. It also records and transcribes meetings, spotting Zoom, Teams and FaceTime calls on its own and labeling who said what. Transcription can run entirely on local models (Whisper, NVIDIA's Parakeet and others), so your audio never has to leave your machine. MIT licensed, free, no account needed. Install is a prebuilt download for Mac, Windows or Linux, and it updates itself. A GPU speeds it up through Metal, CUDA or Vulkan. You can plug in your own OpenAI, Groq, Gemini, Mistral, Deepgram or AssemblyAI key for unlimited cloud transcription on the free plan. Usage analytics stay off unless you turn them on. Intel Macs miss the live speaker-ID features. The startup behind it sells a cloud on top. Free includes 2,000 cloud words a week and 5 meeting hours a month. Pro is $8 a month ($80 a year) for unlimited cloud transcription, agent mode, sync, the iPhone app and 20 meeting hours. Business is $16 per user per month for unlimited meetings and team admin, and SSO means Enterprise. Solo: the free local app is plenty. Teams leaving Otter: price Business against your current bill. The catch: the desktop app is open source, the cloud backend is not, so sync and the phone app can't be self-hosted. It also moves fast, with about ten releases in two months and a long issue queue, so expect the occasional rough update.

vibe8.0k★

Vibe is another desktop transcription app built on Whisper, and what sets it apart is how much it does after the transcript. Everything runs locally, it handles almost every language with translation to English, and it exports to more formats than most people will ever need: SRT, VTT, TXT, HTML, PDF, JSON, and DOCX. MIT licensed, completely free. It is built on whisper.cpp, the C++ port of Whisper, which tends to be lighter on resources than the Python original. You get batch transcription, speaker identification, microphone and system-audio recording, and a command-line interface plus an HTTP API for wiring it into other tools. The standout feature is AI summaries: point it at Claude or a local Ollama model and it condenses a transcript for you. Solo and small teams: it is a free download on macOS, Windows, and Linux, with mobile on the way. If buzz is the no-frills on-ramp, vibe is for people who want the transcript to feed straight into a workflow. The export options and the HTTP API are the reason to pick it. The catch is that the AI summary feature leans on an outside model, so the local-only promise holds for transcription but not for summaries unless you run Ollama yourself. And like every Whisper wrapper, accuracy is only as good as the model and the hardware you give it.

SmartSub6.0k★

SmartSub takes audio or video and produces translated subtitles, start to finish, on your own machine. It transcribes speech to text, translates it, lets you proofread against a video preview, then burns the subtitles into the video or muxes them as a soft track. Nothing gets uploaded, which matters if you're working with footage you can't send to a cloud service. MIT-licensed and free on Windows, macOS, and Linux. It ships six transcription engines, including whisper.cpp and faster-whisper that run fully offline, and hooks into 17+ translation providers from Google and Baidu to Gemini, DeepSeek, and any OpenAI-style API. GPU acceleration covers NVIDIA CUDA, AMD and Intel via Vulkan, and Apple Metal, with automatic detection and CPU fallback. Setup is heavier than a pure web tool: you download models, pick an engine, and configure providers, but it's a desktop install, not a server. Solo creators and translators: this is a strong free alternative to paid subtitle services, and the local engines cost nothing to run. Small teams doing video work: same, with no per-seat fee. Large teams: workable, though there's no central account or shared workflow. The catch: "free" stops at the app boundary. The bundled offline engines are free, but the moment you wire up Gemini, DeepSeek, or a cloud translation API for better quality, you're paying those vendors per use on your own keys. Stick to whisper.cpp and faster-whisper and it's genuinely free end to end.

Auto-Editor5.8k★

Auto-Editor cuts the dead air out of a video in one command. Point it at a file and it measures loudness across the timeline, finds the silent stretches, and removes them. That is the most tedious pass in any talking-head or gameplay edit. Public domain under the Unlicense. No server, no GUI, no dependency tree. Grab the binary from GitHub releases, Homebrew, or the Arch AUR. Silence is not the only trigger: motion detection removes still frames, and margin settings keep the cuts from feeling clipped. It renders the result or hands off a timeline to Premiere, Final Cut, Resolve, Kdenlive, or ShotCut. Solo creators should install it today. Teams can drop it into a render pipeline since it is a CLI. It does not transcribe, so it cannot cut filler words by name the way Descript or wassgha/rescript can. It only knows loud from quiet. The catch is installation. Auto-Editor dropped its Python packaging at version 30, so the newest release on PyPI is two majors behind and anyone who installed with pip is silently running old code. The paid GUI at app.auto-editor.com is a separate product from the same author, one-time from $23.99, commercial use only at $179.

AI-Youtube-Shorts-Generator5.8k★

AI-Youtube-Shorts-Generator turns a long YouTube video into vertical shorts. It transcribes the video, has an LLM score the moments most likely to hold attention, and crops each one to 9:16. It's pitched as the open source answer to OpusClip and Klap, and the code is MIT. Two modes, and the difference is the whole story. The default API mode runs every step through MuAPI, a paid API. Local mode does download, transcription and cropping on your machine and only calls OpenAI or Gemini to rank clips. Gemini's free tier covers that, so local mode can cost nothing. It's a Python CLI: no editor, no preview, no captions. Solo creators: local mode with a Gemini key, free. Small teams clipping at volume: API mode costs cents per clip. Agencies that need captions and an editor should pay for a SaaS clipper. The catch: this repo is a storefront. MuAPI is run by Vadoo, the company the repo's homepage points to, and the README carries tracking links to MuAPI's paid AI Clipping API, which it says makes better clips. No code has changed since June; every commit since touched only the README or license. Treat it as a scaffold.

soundcraft4.1k★

SoundCraft is a clean-room rebuild of Avid Pro Tools in Rust: the Edit and Mix windows, edit modes, playlists, automation, a MIDI piano roll, and its own EQ, compressor, reverbs and synths. It's free under MIT or Apache-2.0, with builds for macOS, Windows, Linux, FreeBSD and the browser. Install a release and open the demo song. Your CLAP, VST3 and Audio Units plugins load, but not AAX, Pro Tools' own format, and the maintainers call AAX likely never. On macOS, the signed builds currently refuse third-party VST3 (an open issue). An MCP server lets Claude run a mix and bounce it headless. There's no paid tier. Solo: free, fine for cutting a podcast or sketching a song. Small studios and post houses: stay on Pro Tools, or try Ardour, the mature open source DAW. Larger facilities: not close. The catch: its own roadmap calls it pre-alpha, about 40% of the way to real work, and it went public a week ago, built at speed by AI coding agents. You can't record MIDI from a keyboard yet, and there's no AAF, OMF or .ptx, so a session someone sends you won't open. Five platforms build; only macOS has been exercised.

flowblade3.1k★

Flowblade is a desktop video editor for Linux: cut clips on a multi-track timeline, add filters and titles, and export. Janne Liljeblad has led it for over a decade, and it is GPL-3.0 and free with no paid edition. It is for cutting footage by hand, not an AI clip generator. Install is a Flatpak from Flathub or your distro's package manager, so there is nothing to host. It runs on the MLT engine, the same one behind Shotcut and Kdenlive. You get up to 21 video and audio tracks, 50+ image filters, 30+ audio filters, keyframed compositing, batch rendering, and NVENC or VAAPI hardware encoding. Solo creators and small teams on Linux can edit real projects with it and pay nothing. CapCut has no Linux desktop app, so for a Linux shop this is a practical replacement for the editing part, just not the templates or auto-captions. OpenCut-app/OpenCut is the browser-based option if you want something closer to CapCut's feel. The catch is scope and platform. Linux only, no Mac or Windows build. The known-issues list says MKV recordings from OBS may not edit properly until you re-encode, and several long-standing bugs are marked as not planned for a fix.

SnapOtter2.9k★

SnapOtter converts, compresses, OCRs, and transcribes almost any file you hand it, on your own hardware. Images, video, audio, PDFs, spreadsheets, through a web UI, a REST API, or saved pipelines. Local AI covers background removal, upscaling, face blur, and transcription with no outside calls. The AGPLv3 build includes all 200-plus tools. One Docker command runs it, Postgres and Redis baked into the image; production gets a three-container compose stack. Two cores and 2 GB of RAM starts it, the full AI model packs want 4 cores, 4 GB, and about 20 GB more disk. GPU acceleration is NVIDIA CUDA only, so Intel and AMD integrated graphics buy you nothing. Solo and homelab: run the free image and close the converter tabs. Small teams: OIDC login, API keys, batch, and pipelines are all in the community build. Medium and large: SAML, SCIM, MFA, and audit export need a license key with a seat count and no published price. One job only? tesseract-ocr/tesseract, openai/whisper, and FFmpeg/FFmpeg are lighter. The catch is S3. Object storage sits behind the same license key as SAML and SCIM, so the free build keeps every file on the container disk. Gating auth and compliance is fair. Charging to store files anywhere but local disk, on a product whose whole job is files, is not.

OpenChatCut2.9k★

OpenChatCut puts a conversational agent and a real multitrack timeline in the same window. Tell it what you want and it writes actual clips, transitions, captions and effects into the project, so you keep editing by hand afterward instead of accepting whatever a generator handed back. It is AGPL-3.0 and everything stays on your machine. Desktop installers cover macOS, Windows and Linux, but they are early builds and the macOS packages are unsigned, so the operating system will fight you on first launch. Running from source needs Node 24. The timeline and manual editing work with zero cloud configured. Transcription and generation need your own provider keys. Bring your own keys is the whole pricing story. The commercial ChatCut sells credits starting at $25 a month for 100 of them, and credits are what buy generation time. OpenChatCut charges nothing and sends you to the provider at cost. Solo creators and small teams come out ahead. The catch is the license. AGPL-3.0 means wrapping this in a hosted service obligates you to hand users the source. For editing on your own desktop, irrelevant. For anyone planning to build a product on top of it, decisive.

story-to-handdrawn-video2.4k★

Feed this a short story, or a set of ordered images, and it renders a vertical hand-drawn "diary comic" video with handwritten captions, then hands you a silent MP4 to drop a voiceover onto later. It's built on Remotion, and it ships both the renderer and an agent-skill wrapper so you can drive the whole thing in plain language. Under the hood it wants Node 20+, Python, FFmpeg, and Chrome, plus your own OpenAI API key for the copy and image steps. So the tool is free but the generation isn't: you pay OpenAI for whatever it calls. Setup is more involved than a one-liner, which is why this leans moderate rather than trivial. For solo creators and small teams it costs nothing beyond the API usage. Reach for it when you want a specific aesthetic, the hand-drawn storybook look, without hiring an animator. The catch is that this is a novelty with a narrow lane. It's oriented around Chinese story copy, the output style is fixed, and it solves exactly one creative problem. Impressive to watch, but don't expect a general video pipeline.

JIZURA2.2k★

JIZURA turns song lyrics into a lyric motion video right in your browser. Paste the lyrics, load the song, and it assembles animated typography from hundreds of small layout, motion and effect pieces; a new random seed gives you a new arrangement. Beat detection and tap-to-sync line the words up with the music. MIT, free, nothing to install. Everything runs client-side, so your lyrics and audio never leave the browser. Export an MP4 with the song, PNG sequences, transparent layers or green-screen backgrounds, anywhere from 720p to 4K. MP4 export needs a browser with WebCodecs, and Chrome or Edge is the safe choice. There's also an After Effects export for finishing work. Musicians and anyone making lyric videos for YouTube or TikTok: this is a fast, free first cut that replaces an afternoon of keyframing. The styles lean Japanese in fonts and typesetting, though Chinese and Korean lyrics work too. The catch is the optional After Effects panel installer. It turns on Adobe's PlayerDebugMode, which lets After Effects load any unsigned panel, not just this one. The browser app needs none of that.

baocut1.7k★

BaoCut is a desktop video editor you drive by asking. It transcribes, subtitles, translates, dubs, trims filler out of talking-head recordings, and every result lands on a timeline you can keep editing by hand. It covers a lot of what people pay Descript, CapCut and OpusClip for, on Apple Silicon Macs and Windows, and it's free. No server, but it doesn't bring its own brain: agent features need a signed-in Claude Code or Codex CLI, and you install ffmpeg yourself. Transcription and speech can run on local models or call OpenAI, ElevenLabs and a dozen more, with a ledger tracking cost per job. Your own agent can drive it over CLI or MCP. There's no paid tier. The real bill is your agent subscription plus any API calls. Solo creators: free, and you can sell what you make. Small teams: free for internal use. Larger shops: same, until you want to resell it, host it or embed it, which needs a commercial license. The catch: it's source-available, not open source, under a custom license since October 2026, and some local voice and image models are non-commercial only. Windows builds are unsigned. Three months old, several releases a week: expect churn.

ffmpeg-normalize1.5k★

ffmpeg-normalize fixes audio that is too loud, too quiet, or all over the place from clip to clip. Point it at a file or a whole folder and it normalizes loudness to broadcast-style targets (EBU R128) or simple peak and RMS levels, using FFmpeg under the hood. If you have ever stitched together video where one clip blows out your speakers and the next is a whisper, this is the tool that makes them match. It is free, MIT-licensed, and it is both a command-line tool and a Python module. There is almost nothing to set up beyond having FFmpeg installed; it is a pip install or a Docker image, and it is maintained by Werner Robitza, who is steady and responsive about it. The one thing to understand is that loudness normalization is opinionated: EBU R128 targets a standard that is right for spoken-word and broadcast but can feel flat for music, so the defaults will not suit every job. The flags to override that are all there once you read the docs. For anyone doing podcasts, video editing, or batch audio cleanup, this belongs in your toolbox. There is no paid version and no reason for one. The alternatives are doing it by hand in an audio editor, which is slower, or wiring up raw FFmpeg loudnorm filters yourself, which this tool exists to save you from.

pireel1.4k★

Pireel is a video editor for talking-head content, the kind of thing you'd make for YouTube or a product demo, and it runs entirely in your browser. Timeline, animated captions, themes, graphics, live preview, export, no account and no server for the core editing. The editor is open source under AGPL and free. Running the editor yourself is easy, it's a browser app. The AI features are where it gets real. Narration, transcription, image and video generation, cross-device sync: those aren't self-contained. You either wire in your own AI providers and backend, or you use Pireel's hosted service. So "self-hosted AI video editor" is half true. The editing self-hosts, the AI doesn't. Hosted pricing runs free for 20 credits a day, $17-20/mo for Pro with actual video generation, and up to $83-100/mo for the heavy tiers, plus credit top-ups. Solo creators who just need to cut and caption talking-head video can self-host the editor and pay nothing. If you want the generative features without standing up a backend, the paid cloud is the honest path. Descript, Veed, and Submagic are the paid tools this is aimed at. The catch is that split between the open editor and the paid AI, with AGPL on top. The editor is free and capable, but the "AI" in the pitch lives behind either your own infrastructure or Pireel's subscription, and AGPL makes commercial forks a legal question, not a copy-paste. Know which half you're getting.

aTrain1.2k★

aTrain turns speech recordings into text on your own machine, with no cloud upload and no subscription. It runs OpenAI's Whisper model locally for transcription in 99 languages and adds speaker diarization (working out who said what) through pyannote. It is a real desktop app with installers on the Microsoft Store and Flathub, not a script you have to babysit. AGPL-3.0, fully free. Because everything runs on your device, nothing you record leaves your computer, which is the whole point for anyone handling interview, medical, or legal audio. On a plain CPU it is slow; with an NVIDIA GPU and the CUDA toolkit installed, the best model runs at roughly three times the audio length. It exports straight into MAXQDA, ATLAS.ti, and NVivo, so qualitative researchers are clearly the target audience. Weigh this against Otter.ai, Rev, and Trint, which are faster and need no setup but send your audio to their servers and bill you monthly. If privacy matters, or you transcribe enough hours that subscriptions add up, aTrain wins outright. Solo researchers and journalists can install it and stop paying per minute. Teams with sensitive recordings may find local processing is the only option compliance allows. The catch: local means your hardware is the bottleneck. Without a decent GPU, long recordings take real time, and accuracy still depends on audio quality the way every transcription tool does. It trades a monthly bill for your own patience and a CUDA install.

Rescript1.0k★

Rescript deletes video the way you delete text. Drop in a recording and Whisper transcribes it on your own machine with word-level timestamps and speaker labels. Delete a sentence from the transcript and the footage is cut. That is Descript's defining trick, with no account and no subscription. Free desktop builds for macOS, Windows, and Linux, plus a browser version. Nothing to host. Transcription, speaker detection, and export all run in-process through transformers.js and ffmpeg.wasm, so your laptop is the whole stack. The Whisper model downloads once, then you can pull the network cable. Filler and silence removal are one click, and it exports a finished cut or a timeline for DaVinci Resolve, Premiere, Final Cut, and AAF. Solo creators cutting podcasts and talking-head video are the fit. Small teams can pass files around but not projects: no collaboration, no cloud, no screen recorder. Anyone needing a review workflow should stay on Descript or look at pireel/pireel. Three catches. Rescript left MIT for the PolyForm Noncommercial license, so paid client work needs a separate deal with the author. The voice regeneration the marketing site leads with sits under Next on the project's own roadmap, not in the code. And the public repo is weeks old with one maintainer, so treat it as promising, not proven.

jzsub1.0k★

jzsub takes a video link and hands back a finished MP4: highest quality available, cover art, and bilingual subtitles burned in, translated by GPT. It works across YouTube, Bilibili, TikTok, Vimeo, and anything else yt-dlp can reach. MIT licensed and free; you bring your own LLM key for the translation step. The setup is the tax. Python 3.10+, yt-dlp, Deno, and an FFmpeg build compiled with libass for subtitle rendering, plus a specific font. That FFmpeg requirement is what trips people up. The docs are Chinese-first, so English speakers are translating as they go. Use it if you archive or localize video with dual-language captions and want one command instead of a four-tool pipeline. For plain downloads, yt-dlp alone is simpler. For transcription without the download step, Whisper-based tools do the job. The catch: this is one maintainer's pipeline, tuned to their workflow. If your FFmpeg build or font doesn't match, you're debugging the plumbing yourself.