Video Summarizer
summarize.martino.im
- Free plan
- Windows
- macOS
- Linux
- Web
- API
- Self-hosted
What Video Summarizer is
Video Summarizer is a free, open-source tool that turns video URLs or files into reusable notes. It accepts YouTube, Instagram, TikTok, Twitter/X, Reddit, Facebook, Google Drive, Dropbox, and local files. Its default workflow uses captions when available or transcribes audio with Whisper, then summarizes the transcript with an LLM. A visual mode instead sends video, visible content, and audio directly to a video-capable model. Users can choose Q&A, distillation, fact-checking, tutorials, Mermaid diagrams, essays, or custom styles through prompts.json. Interfaces include a Streamlit GUI, CLI, optional HTTP API, Docker deployment, Raycast extension, and agent skill. The project is local-first and self-hosted, requiring Python 3.7 or later, ffmpeg on PATH, and at least one LLM API key in a local .env file. Transcript caching supports reruns with other styles or providers. Public Tinypaste sharing creates pages visible to anyone with access.
Who it suits
It suits people condensing podcasts, lectures, conference talks, and social videos into notes. Developers and self-hosting users can choose local interfaces, APIs, providers, and custom prompts.
Good to know, and what to watch
- Free and open source
- Supports many video and file sources
- Transcript and visual processing modes
- Multiple interfaces, including CLI and Docker
- Transcript caching avoids repeated transcription
- Requires an LLM API key
- Requires Python and ffmpeg
- Visual requests default to 120 seconds
- Shared summaries are public pages
The verdict
Video Summarizer offers broad input support and configurable note formats across local and hosted workflows. Its requirements and public sharing behavior matter for anyone planning private or controlled processing.
What the maker publishes
- Bring your own key
- Users supply at least one LLM API key in a local .env file; configuration is stored in summarizer.yaml and .env. source
- Caching
- Transcript caching lets users rerun a video with different styles or providers without transcribing it again. source
- Input sources
- It supports YouTube, Instagram, TikTok, Twitter/X, Reddit, Facebook, Google Drive, Dropbox, and local files. source
- Integration
- The Raycast extension can send a video URL to the local Summarize API for summarization. source
- Interfaces
- It provides a Streamlit GUI, CLI, optional HTTP API, Docker deployment, Raycast extension, and agent skill. source
- Interfaces
- The maker documents a CLI, Streamlit GUI, HTTP API, Docker deployment, Raycast extension, and agent skill. source
- LLM integrations
- It works with OpenAI-compatible endpoints, with examples including OpenAI, Groq, Gemini, Ollama, OpenRouter, NVIDIA, and Perplexity. source
- LLM providers
- The maker lists OpenAI, Groq, Gemini, Ollama, OpenRouter, NVIDIA, Perplexity, LiteLLM, and any OpenAI-compatible endpoint. source
- Maker
- The maker's personal site identifies Martino Piaggi as a computer science engineer who primarily writes open-source code. source
- Notable sharing limit
- Shared summaries are public pages, according to the sharing documentation. source
- Prefiltering
- Optional JEV prefiltering selects transcript passages before summarization and can include or exclude specified concepts. source
- Prefiltering
- Optional JEV prefiltering selects transcript passages for inclusion or exclusion before sending text to the summary model. source
- Processing modes
- It offers transcript based summarization and a visual mode that sends video directly to a vision-capable model without transcription. source
- Requirements
- The documented requirements include Python 3.7 or later, ffmpeg on PATH, and at least one LLM API key. source
- Security and compliance
- The maker describes the project as self-hosted and local-first and instructs users to keep API keys in .env; the opened maker pages state no security certification or compliance standard. source
- Security and data handling
- The project describes itself as local-first, and users configure provider API keys in a local .env file. source
- Sharing
- The webapp can publish a summary as a public rendered page through Tinypaste without an account or token. source
- Summary formats
- Available summary styles include Q&A, distillation, fact-checking, tutorials, Mermaid diagrams, and essays, with custom styles added through prompts.json. source
- Summary styles
- Built-in styles include Q&A, distillation, fact-checking, tutorials, Mermaid diagrams, and essays, with custom prompts supported through prompts.json. source
- Supported sources
- The maker lists YouTube, Instagram, TikTok, Twitter/X, Reddit, Facebook, Google Drive, Dropbox, and local files as supported sources. source
- Transcript cache
- A transcript cache allows rerunning the same source with different styles, providers, or languages without transcribing it again; disk persistence can be enabled. source
- Transcript workflow
- The default workflow uses captions when available or transcribes audio with Whisper, then summarizes the transcript with an LLM. source
- Visual mode
- Visual mode sends video, including visible content and audio, to a video-capable model without transcribing first. source
- Visual mode limits
- The documented default visual limits are 120 seconds per request and 100 MB per file, with MP4, MPEG, MOV, and WEBM listed as supported formats. source
Checked against the maker's pages on 9 Oct 2026.