PlugWiki

Can Claude watch videos? What works, and which AI can

Claude cannot watch videos yet. Gemini can, and Claude can still work from a transcript, screenshots or a plug that turns a video into frames.

Picks
5
ranked

No. Claude cannot open a video file or a YouTube link and watch it, in the app or in Claude Code. It reads text and images, so you can give it a transcript or still frames, or add a plug that makes them for you. If you want an AI that takes the video itself, use Gemini.

Good for

  • People who want a YouTube video, lecture or talk summarized
  • Developers who want Claude Code to see a screen recording of a bug
  • Creators and marketers reviewing clips, ads or reels
  • Teams turning recorded demos and calls into notes

The short answer

Use Gemini when the AI has to watch the video itself: it takes uploads up to 5 minutes without a paid plan and 1 hour on Google AI Pro. Stay in Claude for transcripts and screenshots, and add claude-real-video in Claude Code to have it pull the frames for you.

Our pick: Gemini

Every route outside Gemini works from still frames or text, so Claude can miss motion and anything that is shown but not said.

Best for

  • Long videos and YouTube links: Gemini
  • Summarizing a talk or tutorial: paste the transcript into Claude
  • Bug recordings in Claude Code: claude-real-video

Skip it if

  • You need Claude to follow what was said in a video with no captions and no transcript
  • Your footage is private and cannot be sent to an AI provider, even as frames

Summarize a YouTube video with Claude

Claude

  1. 1

    Copy the transcript from YouTube

    Open the video on youtube.com in a browser. In the video description, click Show transcript, then select the text in the transcript panel and copy it.

    Tip: YouTube shows a transcript only for videos that have captions.

  2. 2

    Paste it into Claude and say what you want

    Open a new chat on claude.ai or in the Claude app and paste the transcript. Claude cannot open the YouTube link itself, so the text is what it works from.

    Prompt
    Summarize this transcript in 8 bullets with timestamps, then list every tool or product the speaker mentions.
  3. 3

    Ask follow-up questions

    Keep going in the same chat. Ask for steps, quotes or a short post, and ask Claude to flag anything that seems to depend on what was shown on screen.

    Prompt
    From this transcript, write the steps the speaker demonstrates as a numbered list, with the timestamp of each step.

Show Claude what happens in a screen recording

Claude

  1. 1

    Turn the recording into still images

    Claude reads images, not video. Take screenshots at the moments that matter, or, if you have ffmpeg installed, save one frame per second with ffmpeg -i recording.mp4 -vf fps=1 frame_%03d.png.

  2. 2

    Attach the frames in order

    On claude.ai you can attach up to 20 images in one message, so pick the frames around the problem. Skip animated GIFs, because Claude only sees their first frame.

  3. 3

    Tell Claude what the frames are

    Say how far apart the frames are and what you are looking for, so it can follow the order.

    Prompt
    These are frames from a 40-second screen recording, in order, one per second. Describe what the user clicks and where the error first appears.

Let Claude Code watch a video with claude-real-video

Claude

  1. 1

    Install ffmpeg

    The tool uses ffmpeg to pull frames and audio. On a Mac run brew install ffmpeg, on Linux sudo apt install ffmpeg, on Windows winget install Gyan.FFmpeg. You also need Python 3.10 or newer.

    Uses claude-real-video

  2. 2

    Install claude-real-video

    Install it with the transcript option, then copy its skill into your Claude Code skills folder. Run these in a terminal, one at a time:

    • pip install "claude-real-video[whisper]"
    • git clone https://github.com/huangchihhungleo/claude-real-video
    • mkdir -p ~/.claude/skills && cp -r claude-real-video/skills/claude-real-video ~/.claude/skills/

    Uses claude-real-video

  3. 3

    Paste a link or file and ask

    Start Claude Code and paste a video link or a file path with your question. It keeps the frames that change, writes a transcript on your computer, and reads both.

    Prompt
    Watch https://www.youtube.com/watch?v=VIDEO_ID and list the pricing tiers shown on screen, with the timestamp of each.

    Tip: The frames still go to Anthropic when Claude reads them, so keep private footage out.

    Uses claude-real-video

Upload a video to Gemini

Gemini

  1. 1

    Add the video

    Go to gemini.google.com, click Add files, then Upload Files, and pick your video. Each video can be up to 2 GB, and you can add up to 10 files.

  2. 2

    Ask your question and submit

    Type what you want and click Submit.

    Prompt
    Summarize this video and list the timestamps where each new topic starts.

    Tip: Without a paid plan the videos in one prompt can total 5 minutes. Google AI Pro and Ultra raise that to 1 hour.

  3. 3

    For YouTube, ask Gemini directly

    Gemini can find YouTube videos and answer questions about what is in them. You need to be signed in with Keep Activity on.

    Prompt
    Find a short video on getting grape juice out of a wool rug, then summarize the steps it shows.

Best plugs for this

Ranked by GitHub stars, then weekly downloads; the hosted Apify option is listed for people who cannot install anything. None is made by Anthropic.

  1. 1claude-real-video logo
    claude-real-videoMost starredLocalNo sign-in

    Pulls the frames that change and a transcript from a link or file on your computer, then hands them to Claude Code or any AI.

    2,275 GitHub stars
  2. 2claude-video-vision logo
    claude-video-visionLocalNo sign-in

    A Claude Code plugin that extracts frames and analyzes the audio, so Claude can answer questions about a video.

    1,443 GitHub stars
  3. 3mcp-youtube-transcript logo
    mcp-youtube-transcriptLocal

    Fetches YouTube transcripts and video details for Claude. YouTube blocks it from cloud servers, so run it on your own computer.

    494 GitHub stars
  4. 4mcp-video-analyzer logo
    mcp-video-analyzerLocal

    Turns YouTube, Loom, TikTok and other links or local files into transcripts, key frames and on-screen text.

    107 GitHub stars
  5. 5YouTube Transcript (Apify) logo
    YouTube Transcript (Apify)OfficialHosted

    Runs on Apify's servers, so it works in the Claude app with nothing to install. Apify charges per transcript.

Claude

Plan: No Claude plan adds video. Any plan, including Free, works for transcripts and screenshots.

What it can do

  • Summarize, quote or turn a transcript you paste into steps
  • Look at up to 20 images per message on claude.ai, such as frames from a video
  • Use plugs in Claude Code that cut a video into frames and a transcript

Limits

  • It cannot take MP4, MOV or other video files
  • It cannot open a YouTube link and watch it
  • Animated GIFs show only their first frame
  • Frames miss motion between them and use far more of your limit than text

Plugs to turn on

Gemini

Plan: Free covers videos up to 5 minutes in total per prompt. Google AI Pro or Ultra raises that to 1 hour.

What it can do

  • Watch video files you upload, up to 2 GB each
  • Take up to 10 files in one prompt
  • Find YouTube videos and answer questions about what is in them

Limits

  • YouTube answers need you signed in with Keep Activity on
  • YouTube help does not work in Gemini Live or Gemini in Google Messages
  • It cannot see your YouTube watch history, comment or save videos

ChatGPT

What it can do

  • Pull still frames from an attached video with its code tool and describe them, as users report

Limits

  • Frame-by-frame reading can miss what is said in the video
  • Users on OpenAI's forum are still asking for full video and audio support

Microsoft Copilot

What it can do

  • Summarize a YouTube video open in Microsoft Edge, with key takeaways

Limits

  • Chat uploads take documents, images and text files, not video
  • Uploaded files are kept for up to 18 months

Comparison

Use Gemini when the AI has to see the video itself. Use Claude for transcripts and screenshots, or add a plug in Claude Code that makes them for you.

OptionVideo filesYouTube linksLength limitWith a plug
GeminiThis pageYes, 2 GBYes5 min, 1 h paidNot needed
ClaudeNoNoNot supportedFrames + transcript
ChatGPTNot documentedNot documentedNot documentedNot needed
Microsoft CopilotNoIn EdgeNot documentedNot needed

Questions

Can Claude watch videos?

No. Claude cannot open a video file or a YouTube link, in the app, the API or Claude Code. It reads text and images, so give it the transcript or screenshots, or use Gemini for the video itself.

Can I upload an MP4 or MOV file to Claude?

No. Claude's uploads take documents and images, not video or audio. Save a few frames as PNG or JPEG and attach those instead.

Can Claude summarize a YouTube video if I paste the link?

Not by itself. Copy the transcript from YouTube's Show transcript panel and paste it, or add a YouTube transcript plug such as mcp-youtube-transcript.

Can Claude Code watch a video?

Not natively, and Anthropic's feature request for it is still open. Community tools such as claude-real-video and claude-video-vision cut the video into frames and a transcript that Claude Code can read.

Which AI can actually watch videos?

Gemini. You can upload video files up to 2 GB each, and it can answer questions about YouTube videos. Without a paid plan the total length per prompt is 5 minutes, and Google AI Pro or Ultra raises it to 1 hour.

Can ChatGPT watch videos?

Partly. Users report it pulls still frames from an attached video with its code tool and describes them, which misses what is said. Users on OpenAI's forum are still asking for full video and audio support.

Can Microsoft Copilot summarize a YouTube video?

Yes, in Microsoft Edge. Open the video and ask Copilot for a summary and key takeaways. Copilot's chat uploads do not take video files.

Is there an official plug that lets Claude search my videos?

Yes. TwelveLabs runs a hosted connector called Jockey that lets Claude search and ask questions across a video library you upload to TwelveLabs. You sign in with a TwelveLabs account, and a free account is available.

Is it private to let Claude look at my video?

Only partly. Tools like claude-real-video process the video on your computer, but every frame and transcript Claude reads is still sent to Anthropic.

Why does watching a video through Claude use so much of my limit?

Each frame is an image, and a few dozen images cost far more than a transcript. Send fewer frames around the moment you care about, or use Gemini for long videos.

Sources

Docs, issue trackers and discussions this page is built from.

Show all 16
  1. Upload & analyze files in Gemini Apps · support.google.com
  2. Use YouTube in Gemini Apps · support.google.com
  3. Upload files to Claude · support.claude.com
  4. Vision · platform.claude.com
  5. Claude-real-video - any LLM can watch a video · news.ycombinator.com
  6. File upload in Microsoft Copilot · support.microsoft.com
  7. Video highlights in Microsoft Edge · explore.microsoft.com
  8. [FEATURE] Video Input Support in Claude Code · github.com
  9. claude-real-video README · github.com
  10. Claude-real-video - any LLM can watch a video · news.ycombinator.com
  11. View video transcripts · support.google.com
  12. Claude-real-video - any LLM can watch a video · news.ycombinator.com
  13. mcp-youtube-transcript README · github.com
  14. New ChatGPT MP3 and MP4 file support · community.openai.com
  15. Model Context Protocol · docs.twelvelabs.io
  16. Claude-real-video - any LLM can watch a video · news.ycombinator.com

Updated Nov 1, 2026.