Can Claude watch videos?
Last updated 2026-10-20
The short answer
Not as video. Anthropic's help page on uploading files to Claude lists documents and images, and video isn't one of the types. Claude can read still images and text, so the workaround is to give it frames and a transcript. Woven Video, a Mac editor, does that for an assistant: stills of your footage plus a transcript with word timings, but no motion or sound.
Your options
Upload frames or screenshots. Claude reads JPEG, PNG, GIF and WebP images, and a claude.ai chat takes up to 20 files. Pick the frames that matter: the opening shot, each cut, any on-screen text and the last frame. In a 60-second clip, 20 frames means one every three seconds.
Don't use a GIF as a stand-in. Anthropic's vision docs say animations are unsupported and only the first frame is used, so a GIF of your clip shows Claude one still.
Paste a transcript. It tells Claude what was said, not what was on screen. Use it together with frames when the question is about the script, the hook or where to cut.
Use a tool that extracts frames. Claude Code can run shell commands and read images you point it to, so it can run a frame extractor such as FFmpeg, if you have it installed, and then look at the results. You manage the files and the commands yourself.
Connect an app that does it for you. That's Woven Video, below.
Let Claude look at your footage through Woven
Woven Video is an AI video editor for Mac. It hands an assistant the same kind of material you would: still frames, a transcript with word timings and a song match. It never gives motion or sound, and Woven's rule for the assistant is to say so instead of guessing.
- In Woven's own chat. Paste a link to a reel you like and Woven's own agent, not Claude, downloads it, pulls frames, transcribes the speech with word timings, identifies the song and writes up the hook, format, pacing, shots and sound. For longer videos the breakdown covers the first 90 seconds. Login walls, rate limits and geo-blocks can stop a download.
- From Claude Code. Once Claude Code is connected, it can download a video from a link, ask Woven for one to six stills of it at a time (look_at_media), transcribe it, pull out the audio and identify the song. It doesn't get the in-app agent's frame-by-frame extraction or contact sheets.
- Check the edit, not just the source. Claude can ask for one to six stills of the finished reel (render_frame), drawn by Woven's own renderer with pictures, text, captions and effects. The connector won't export until Claude has rendered the current visuals.
- Transcribe once. Transcribing a file once makes later transcription and captioning of that file free for those seconds. Transcription uses Woven credits.
Common mistake. Asking Claude whether the pacing or the music works. Stills show no motion or sound, so ask about framing, on-screen text and the words instead, and use the word timings to judge pacing.
How to let Claude see a video through Woven
- 1Connect. In Woven, open Settings, then AI apps, pick Claude and click Copy setup prompt. Paste it into Claude Code on your Mac.
- 2Load the tools. Woven tells you when it is connected. In a terminal, start a new session in Claude Code to load Woven's tools. In a desktop app, use the restart button Woven offers.
- 3Ask Claude to look. Bring in a file or give it a link, then ask what is in it: the hook, the on-screen text, what is said at each cut.
- 4Check the result. After an edit, ask Claude to render frames of the reel and compare them with the source.
Needs macOS 15 or later, a signed-in Woven account with a plan or the 3-day free trial, and an app that runs commands on your Mac, such as Claude Code. The setup prompt won't run in Claude's chat or in cloud or web sessions.
FAQ
Not as video files. It analyzes still images and text, so the workable route is frames plus a transcript. Through Woven's connector, Claude Code can request stills of a video, a transcript with word timings and a song match, then analyze those.
Anthropic's upload help page lists document and image types, and video isn't among them. Chat uploads allow up to 20 files, and images can be up to 8000 by 8000 pixels.
It sees images. A video has to be turned into frames first, and anything in the sound has to arrive as text.
Not by watching it. Anthropic describes its API web fetch tool as retrieving web pages and PDF documents. Paste a transcript instead, or connect Claude Code to Woven, which can download the video and hand Claude stills, a transcript with word timings and a song match.
Claude Code's docs describe adding images by drag, paste or file path, and don't mention video. With Woven connected, it can ask for stills of a video and a transcript.
Not directly. Anthropic's upload page lists no audio types. Woven turns speech into text with word timings and can identify a song. Nothing else about the sound reaches the assistant.
More answers
Can Claude or ChatGPT edit videos? See our answers for Can Claude edit videos?, Editing with Claude Code or Codex and Can ChatGPT edit videos?.
Sources
- Upload files to Claude (Claude Help Center), checked October 19, 2026
- Vision (Claude Platform Docs), checked October 19, 2026
- Features overview (Claude Platform Docs), checked October 19, 2026
- Common workflows, Work with images (Claude Code Docs), checked October 19, 2026
- Tools reference (Claude Code Docs), checked October 19, 2026
- About FFmpeg, checked October 19, 2026
- FFmpeg Formats Documentation, image2 muxer, checked October 19, 2026
Try Woven free for 3 days
Then $25/mo billed yearly ($299/yr) or $39/month. Native Mac app for short-form video.
Paste into Claude Code to install and connect Woven.