Resources / Antigravity / Free AI video editing pipeline with Google Antigravity
A

Free AI video editing pipeline with Google Antigravity

Edit reels with Google Antigravity instead of Claude Code. Malayalam captions, motion graphics and the final render, without spending Claude tokens.

Tested 11 Nov 2026Google AntigravityAntigravityIntermediate
Watch on Instagram
Watch the reel firstHow to edit for free without Claude Code

What it does

This is the pipeline I used to edit the reel above. It runs in Google Antigravity, Google’s AI coding agent, so editing doesn’t use up your Claude limits. You give it your raw clip and one instruction file, and it:

  • transcribes your speech with word-level timings, including Malayalam
  • cuts at the quiet points between words, with tiny audio fades so you never hear a click
  • adds karaoke-style captions (Malayalam turned into Manglish)
  • builds motion graphics with HyperFrames (HTML turned into video)
  • checks key frames, then renders the final MP4

Before you start

  • A laptop or PC (Windows, Mac or Linux). Rendering is faster with a good graphics card, but not required.
  • A Google account. Antigravity is free to use. A Google AI Pro plan gives you higher limits.
  • Your raw video in one folder.
  • 30 minutes, once.
Jio users: free Google AI Pro

Jio has offered free Google AI Pro to Unlimited 5G users through the MyJio app (look for the Gemini / Google AI Pro banner and tap claim). The offer and its terms change, so check the MyJio app for what’s available on your plan today.

Setup in 6 steps

  1. Install Google Antigravity

    Download it from the official site, antigravity.google, install it, and sign in with your Google account.

  2. Install FFmpeg, Python and Node.js

    FFmpeg cuts the video, Python runs the transcription and Node.js runs HyperFrames. On Windows, open PowerShell and run:

    winget install Gyan.FFmpeg
    winget install Python.Python.3.12
    winget install OpenJS.NodeJS.LTS
    Check it workedClose PowerShell, open a new one, and run ffmpeg -version, python --version and node -v.

  3. Open your video folder in Antigravity

    Put your raw clip in a new folder, then open that folder in Antigravity (File → Open Folder).

  4. Let Antigravity set up the tools

    Paste this into the Antigravity chat. It installs the transcription model and checks HyperFrames:

    Set up this folder for video editing:
    1. Create a Python virtual environment called .venv-whisper and install faster-whisper, soundfile, numpy and fonttools.
    2. Download the Malayalam speech model adalat-ai/ct2-whisper-medium-ml-rmft for faster-whisper.
    3. Run npx hyperframes --version to check HyperFrames works.
    4. Tell me what is installed and what failed. Don't edit any video yet.
  5. Add the edit instructions

    Create a file called outline.md in the folder and paste the edit plan below into it. This is the same plan I use for my reels.

  6. Run the edit

    In the Antigravity chat, type:

    Read outline.md and edit video

    It shows you the cut plan first. Approve the cuts before it renders. The final video lands in the folder.

The edit plan (outline.md)

You are my video editor. Work in this folder. Show me the plan and wait for my OK before cutting or rendering.

1. Look at the footage: run ffprobe for size, fps and HDR. Grab one frame every few seconds into a contact sheet.
2. Transcribe: convert the audio to 16 kHz mono WAV. Split it every 6-10 s at the quietest point. Transcribe each chunk with faster-whisper and the model adalat-ai/ct2-whisper-medium-ml-rmft, with word timestamps. Save tx.json.
3. Fill the gaps: find speech gaps longer than 1.6 s and re-transcribe just those parts.
4. Plan the cuts: find repeats, false starts and dead air. Cut at the quietest point between words. Show me the cut list and wait for my OK.
5. Cut and conform with ffmpeg: trim, join, add 20-30 ms audio fades at every join, scale to 1080x1920, constant frame rate, keyframe every second (-g 30 or -g 60).
6. Captions: convert Malayalam words to Manglish, max 3 words per caption, max 1.6 s each, with a word-by-word highlight.
7. Build the video with HyperFrames: footage, captions and simple GSAP pop-in graphics for key words.
8. Check: run npx hyperframes check, then snapshot 5 key moments and fix anything cut off or overlapping.
9. Render: npx hyperframes render . --sdr -o final.mp4

Common errors

Captions are wrong or words are missingCheck the transcript before you approve the cut. Standard Whisper large-v3 does badly on Malayalam, so make sure Antigravity used the adalat-ai model. Ask it to re-run step 3 (fill the gaps).
The render freezes or the video stuttersPhone videos use a variable frame rate. Ask Antigravity to redo step 5 with a constant frame rate and a keyframe every second.
The render fails with HDR footage or runs out of disk spaceAlways render with --sdr. The HDR path needs a lot of temporary disk space.
Malayalam text shows as boxesThe font isn’t bundled. Ask Antigravity to copy a Malayalam font (for example Nirmala UI Bold on Windows) into assets/fonts and load it with @font-face.
"ffmpeg is not recognized"Close every PowerShell window and open a new one. If it still fails, restart your PC.