Upload a video and get a 5-bullet AI summary plus a full transcript. Perfect for lectures, meetings, podcasts, and long YouTube videos. Powered by Claude Haiku and Whisper. Free.
You don't always have 45 minutes to watch a recorded meeting or a lecture someone sent you. A summary tells you in 30 seconds whether you need the full thing. This AI video summarizer is a two-stage pipeline: Whisper transcribes the audio, then Claude Haiku reads the transcript and writes a tight 5-bullet summary that covers the main beats without hallucination. You also get the full transcript underneath for the parts you do want to read in detail.
Video-native LLMs exist but they're slow, expensive, and still miss things that are said off-camera. A transcript-first pipeline is faster, cheaper, and more accurate because the LLM sees every spoken word exactly once. The tradeoff is that visual cues (slides, whiteboards, body language) aren't captured — for anything that depends heavily on visuals, you'll want to skim the video itself after reading the summary.
Five bullets, each under 20 words, covering the concrete points and decisions rather than generic platitudes. We instruct the model to avoid hedging language (‘the speaker discusses…’) and prefer action-oriented statements (‘adopts JWT over sessions for mobile parity’). The summary language matches the detected transcript language by default — a Spanish talk produces a Spanish summary — but you can override with the Translate option.
Meeting recordings you missed. Lecture catch-up before an exam. Podcast episodes before deciding to listen in full. YouTube videos longer than 30 minutes. Sales call reviews. Interview shortlisting. For meetings especially, the summary plus the timestamped transcript is enough to cite specific decisions back to the meeting without re-watching.
Claude Haiku 4.5 is used by default because of its strong summarization-to-cost ratio. If Claude is unreachable we fall back to a Workers AI Llama-3.1-8b model automatically.
The 5-bullet format is optimized for skimmability. If you need a long summary, copy the full transcript into Claude, ChatGPT, or Gemini directly with a prompt like ‘give me 20 bullets from this transcript’.
Not in the bullet summary itself, but the full transcript underneath has every segment timestamped — so you can jump to the moment in the video that supports any bullet.
Same as transcription limits: 10 minutes per file free, 4 hours Pro. The longer the video, the more context-rich the summary.
Yes. The summary is generated in the same language as the transcript. Use Translate to get an English summary from a foreign-language video.