Antigravity CLI video guide

Complete Setup and Reference videos first, then the steps for your Digital Story track, then Finishing. Replace every part in [square brackets] before you send a prompt.

Before you start

  • Antigravity CLI is not covered by U-M. Files and text the agent reads go to Google, so an identifiable vulnerable person’s footage or transcript belongs in the offline lane on the Tools page.
  • Your own narration, consented staff, public documents and public-domain footage can go in.
  • Never paste an API key or a password into the agent.
  • The AI rules for the Digital Story apply to every step, including the ban on photorealistic generated people and the label on AI-made images and voices.

Setup

Account. Use a personal Google account, not your U-M account. The free student plan raises your usage limits. A full video uses far more quota than a chat, so claim it before you begin.

Terminal. On Windows, open PowerShell from the Start menu. On a Mac, open Terminal. Paste the install command for your laptop and press Enter.

Install on Windows (PowerShell)

irm https://antigravity.google/cli/install.ps1 | iex

Install on macOS

curl -fsSL https://antigravity.google/cli/install.sh | bash

Close the window, open a new one, and check the version. A version number means the install worked.

Version check

agy --version

ElevenLabs key. Skip this if you will record your own voice. Create a key in your ElevenLabs account, paste your key into the matching command below, then close the window and open a new one.

Key on Windows (PowerShell)

setx ELEVENLABS_API_KEY "your key"

Key on macOS

echo 'export ELEVENLABS_API_KEY="your key"' >> ~/.zshrc

Project folder. Paste these lines one at a time. A new terminal opens in your home folder, which is outside Google Drive and Dropbox; video projects are too large for those folders.

First line

mkdir my-video

Second line

cd my-video

Third line

agy

First launch. Choose a color scheme and a display mode, sign in with your Google account, and confirm that you trust the folder. The sign-in covers the agent and its image generation, so no Gemini API key is needed.

Approvals. The agent asks before it creates a file or uses a command. Read the request, then press Y. Use default mode; Shift+Tab switches modes. Do not use --dangerously-skip-permissions.

Tools. Paste this as your first prompt. Remotion is free for individuals and nonprofits under its license.

Tools and test clip

Check whether Node.js, Python and ffmpeg are installed on this laptop, and install anything missing. Explain each step before you do it. Install OpenAI Whisper so narration can be transcribed on this laptop. Then set up a Remotion project in this folder for a 1280 by 720 video at 30 frames per second, and add the Remotion agent skills with npx skills add remotion-dev/skills. Render a three-second test clip with one line of text to out/test.mp4. Tell me its resolution and length, and tell me how to open it.

Watch the test clip before you go on. If it plays, your laptop is ready.

Commands

Antigravity CLI commands and keys, with the use of each
Command or keyUse
agyStart in the current folder
agy -cContinue the latest conversation in this folder
/planReview a plan before the agent builds; type your request after it
/modelChange the model
/helpSee commands and shortcuts
/exit or Ctrl+DLeave the session
Ctrl+CInterrupt the current response

To come back later, open a terminal, type cd my-video, then agy -c.

Practice files

These filled-in Word files are fictional. Use them to try the steps before you use your own material, and do not post a video made from them as fact.

  • Track A: story of self sheet. Put it in your project folder.
  • Track B: communication brief and sources. Put the brief in your project folder and the sources in a folder called sources.
  • Track C: communication brief and interview transcripts. Put the brief in your project folder and the transcripts in assets/transcripts, then begin at Step C2. These transcripts have no video files, so the clip cuts and interview scenes need your own footage.

Reference videos and style

Do this before you write a script. These videos also count toward your Video notes, due the Tuesday of Week 11.

Step 1. Videos. Choose 3 to 5 videos that look the way you want yours to look, in the form of your track. Start from these, then search YouTube for more.

Step 2. Stills and transcripts. You do not need to download the videos. Pause each one about every 30 seconds and take a screenshot (Windows+Shift+S on Windows, Command+Shift+4 on a Mac), about 10 per video, and save them in references/stills inside your project folder. Under the video, click Show transcript, and paste the text into a file in references/transcripts. Add a file called references/links.txt with each video’s title and link.

Step 3. Style notes.

Style notes

Look at every image in references/stills and read references/links.txt and every file in references/transcripts. Write references/style.md, a working style guide for my video, with these parts: the colors as hex codes, including background, ink and accent; the kinds of fonts used for headlines, body text and labels; what appears at the back, middle and front of a frame; where text and source lines are placed; which kinds of images appear (footage, photos, documents, drawings, charts); how often the picture changes, from the stills’ timing; how long a narration line is, and words per minute, from the transcripts; how things probably move, labeled as a guess because stills cannot show motion. Name which video each point comes from. Then list three things all the videos share and three ways they differ.

Read style.md against the videos, change anything that looks wrong, and delete what you do not want.

Style limits

Add these rules to the top of references/style.md and follow them in every step: References are for study only. No still, clip, logo, character, music or wording from any reference video goes into my video. Do not copy one channel’s look as a set. Combine techniques from several videos and use my own colors and material.

Step 4. Fonts. Choose on Google Fonts, where every font is free to use. Pick one for headlines and one for body text or labels. Google’s short guides are Choosing type and Pairing typefaces. The fonts below are from the class examples. Handwriting fonts suit short labels only. If your video has text in another language, use the Language filter on Google Fonts and check that the font has those characters; the Noto Sans family covers most languages.

Google Fonts pairings by track
TrackHeadlines or labelsBody text or quotations
A, class examplePatrick HandNewsreader
B, class exampleBricolage GrotesqueDM Mono
CThe organization’s own font if it is free to use; otherwise Public SansNewsreader

Font test

Add the package @remotion/google-fonts to the Remotion project, which loads Google Fonts into a video. Make one still at 1920 by 1080 with a headline and two lines of body text from my video, in three pairs of fonts: [font 1 and font 2], [font 3 and font 4], [font 5 and font 6], each pair labeled with its names. Body text must be at least 24 pixels at 1080p. Tell me whether every character displays. When I choose a pair, write it into references/style.md.

Track A, personal story

The class example is a Draw My Life board: white ground, black line drawings that appear stroke by stroke, handwritten labels, and your own photos.

Step A1. Script. Put your completed story of self sheet, the Word file, in your project folder.

Script table, Track A

Read my story of self sheet, the Word file in this folder. Write a first-person script for a video of about [3 minutes] as a table in script.md, with one row per narration line. Columns: number, narration, drawing on screen, photo or document on screen, handwritten label, and source. Use only events, names and words from the sheet. Where a line needs a detail I have not given you, write ASK ME instead of filling it in. Do not invent feelings, quotations or dialogue.

Read the whole script aloud and answer every ASK ME row. Change anything that does not sound like you.

Step A2. Photos and other material. Put your own photos in assets/photos and any screenshots in assets/screens. Anyone identifiable watches the cut before it screens.

Asset research, Track A

List the files I placed in assets, and match each one to a row of script.md. For rows that still need a place, an object or a document, find public-domain or openly licensed material: Wikimedia Commons, Library of Congress, Internet Archive and National Archives. Do not download or use any image of a real person unless I named that person in the script. Save each item in assets. Record the license exactly as the source page states it and never guess one. Write MANIFEST.md with the file, the source URL, the license and the credit line. Tell me which rows have nothing usable. If a site blocks a download, give me the URL and I will save the file myself.

Step A3. Drawings. Simple figures can be drawn in code, so they appear stroke by stroke.

Drawings in code

For each row of script.md that needs a drawing, make a simple black line drawing as an SVG path in the Remotion project: stick figures, a desk, a house, a phone, arrows. Use a marker-style line with round ends and no fill or shading. Save them as reusable components, and show me all of them on one still so I can check them.

For a more detailed illustration, the agent can generate one. Label it as AI-made in the video. Image quota is limited, so generate only what the script needs; if it is used up, send the same description in Gemini with your U-M account and save the file yourself.

Illustration

Generate an image with your image generation tool: [a cartoon person with short hair reading at a desk] as black marker line art on a pure white background. Simple shapes, no shading, no color, no text. It must look like a hand-drawn cartoon, not a photograph. Copy the file to public/drawings and tell me its name.

Step A4. Narration. Use your own voice. Record on your phone in a quiet room, read the whole script once, and put the file in assets/recordings.

Your own recording

Transcribe the file in assets/recordings with Whisper on this laptop. Split it into one audio file per row of script.md, using the transcript to find where each line starts and ends. Trim long silences, even out the loudness, and save the files as public/vo/01.wav, public/vo/02.wav and so on. List any line where the recording differs from the script, and list the length of each file.

To use a clone of your own voice instead, which needs an ElevenLabs Starter plan, or a synthetic voice, use the narration prompt in Track B and label the voice as AI-made in the video.

Step A5. Visual system.

Visual system, Track A

Read references/style.md and follow it. Set up the Remotion project in a Draw My Life style. Every scene has a white background, black ink with one red accent, and handwritten labels in the Google font [Patrick Hand]. Lines draw themselves stroke by stroke. Photos appear as small prints with a white border and a slight tilt. Quotations use the Google font [Newsreader]. Make one test scene and show me a still before you build the others.

Step A6. Scenes. Send this for one scene, check it, and repeat for the others.

One scene, Track A

Build scene [1] from row [1] of script.md. The drawing appears stroke by stroke while the line is spoken, each handwritten label appears at the moment its word is said, and the photo pops in with a small spring. Leave captions out of the picture; they go up as a separate file. The scene lasts as long as its narration file plus two-thirds of a second.

Remotion Studio

Start Remotion Studio and tell me the address to open in my browser.

Look at each scene in the browser and describe what to change in plain words (“make the desk smaller and move it left”). When every scene is done, go to Finishing.

Track B, an explainer

The class example uses a felt board, halftone photo prints, primary documents with highlighted lines, and cut-out buildings in front.

Step B1. Script. Put your communication brief, the Word file with your ABT sentence added, in your project folder, and your sources in a folder called sources.

Script table, Track B

Read my communication brief, the Word file in this folder, and every file in sources. Write a script for a video of about [3 minutes] as a table in script.md, with one row per narration line. Columns: number, narration, what is on screen, image or footage in the middle of the frame, a structure or object in front, and the source to cite on screen. It goes from what the problem looks like, to why it persists, to one thing the audience can do. Use only facts from my sources. Where a line needs a fact I have not given you, write NEEDS SOURCE in the source column instead of filling it in.

Read every NEEDS SOURCE row and every number against the page it came from before you go on.

Step B2. Assets. Read Finding footage and images first, and name the archives that fit your topic in the prompt.

Asset research, Track B

For each row of script.md, find public-domain or openly licensed material I can use: archival footage from [Internet Archive, Library of Congress, National Archives], primary documents as PDFs from [govinfo, Federal Register], and photos from [Wikimedia Commons]. Save items in assets/footage, assets/documents and assets/photos. For footage, download the largest original file, not the default copy. Record the license exactly as the source page states it and never guess one. Write assets/MANIFEST.md with the file, the source URL, the license and the credit line I should print on screen. Tell me which rows have nothing usable. If a site blocks a download, give me the URL and I will save the file myself.

News and organization pages are screenshots, credited “Screenshot of [outlet], [month year]”. A news photo inside a screenshot belongs to the outlet; crop to the headline if the video will be posted publicly.

Step B3. Prints.

Prints

Make each photo in assets/photos a black-and-white halftone print with a white border, about 900 pixels wide, and save it in public/prints. Keep the faces readable. Then show me all the prints on one sheet, each labeled with its file name.

Step B4. Document highlights.

Highlight positions

For each PDF page in assets/documents, find these exact phrases on the page and give me the rectangle of each line they cover, in pixels of a 200 dpi render, so the highlighter falls on the printed words: [paste the phrases]. Save each page as an image in public/docs. Then show me each page with a red box drawn over each rectangle so I can check them. If a phrase is not found, show me the page text around where it should be.

Scanned court opinions often split a word across lines, so a short phrase from one line is easier to find than a whole sentence.

Step B5. Cut-outs. The agent can generate the structure in front of each scene (a building, a desk, a row of chairs). Label these as AI-made in the video. Image quota is limited; if it is used up, send the same description in Gemini with your U-M account and save the file in assets/cutouts yourself.

Cut-out

Generate an image with your image generation tool: [a brick school building] as a flat paper cut-out, seen from the front, on a solid bright green background. Black-and-white with a printed halftone texture. No people, no text, no shadow. Copy the file to assets/cutouts.

Green removal

Remove the green background from each image in assets/cutouts, save transparent PNG files in public/fg, and show me them all on one sheet against a dark background so I can check the edges.

Step B6. Narration. Try a voice first: paste one script line on the ElevenLabs text to speech page, pick a voice from the Voice Library and listen.

Narration

Read my ElevenLabs key from the environment variable ELEVENLABS_API_KEY; do not print it. Make one narration file per row of script.md with the voice [name], and save them as public/vo/01.mp3, public/vo/02.mp3 and so on. Spell numbers out in the text sent to the voice, and leave digits in the script. Tell me how many characters this will use before you begin. Then transcribe each file with Whisper on this laptop, and list any number, name or acronym that was read differently from the script, and the length of each file.

Step B7. Visual system.

Visual system, Track B

Read references/style.md and follow it. Set up the Remotion project so every scene uses the same background texture, the Google font [Bricolage Grotesque] for headlines, [DM Mono] for body text, black ink, one red accent and a yellow highlighter. Each scene has the background, the main image in the middle, and a cut-out from public/fg in front that covers its lower edge. Make one test scene and show me a still before you build the others.

Step B8. Scenes. Send the matching prompt for one scene, check it, and repeat for the others.

One scene, Track B

Build scene [3] from row [3] of script.md. The print pops in with a spring and a slight tilt, the document slides in on the right, and the camera pushes toward the highlighted lines before the highlighter is drawn across them. Put the source line at the lower left and leave captions out of the picture; they go up as a separate file. The scene lasts as long as its narration file plus two-thirds of a second.

Footage scene

Build scene [5] from row [5] of script.md using the clip [file name] from [start time] to [end time]. Show it inside a paper frame with a slight tilt, mute its sound, and print its credit line from assets/MANIFEST.md at the lower left. The scene lasts as long as its narration file plus two-thirds of a second.

Data scene

Build scene [9] from row [9] of script.md using [data.csv]. Show [one small paper figure for each million people / a line that draws from left to right / a state map that changes year by year], with a large counter that matches the number being spoken and the source at the lower left. Use only numbers in the file, and read each number back to me before you build.

Remotion Studio

Start Remotion Studio and tell me the address to open in my browser.

Look at each scene in the browser and describe what to change in plain words. When every scene is done, go to Finishing.

Track C, a video for an organization

The organization agrees to the brief before you film, and anyone identifiable watches the cut before it screens. The class example introduces a website with screen recordings, chalk figures and quote cards; your video can also use footage of the program and interviews with staff or participants.

Step C1. Footage and files. Put video files in assets/footage, and the organization’s logo and photos, used with its permission, in assets/org. Put your communication brief, the Word file with your ABT sentence added, in your project folder. Footage of an identifiable vulnerable person goes to the offline lane.

Footage log

For each video in assets/footage, report its length and resolution, and transcribe it with Whisper on this laptop into a timestamped text file in assets/transcripts. Make one contact sheet per video with a still every 10 seconds, each labeled with its time. Write footage_log.md with a short line on what each video shows. Do not change the original files.

Step C2. Paper edit. Check each transcript against the recording first, names especially.

Paper edit

Read my communication brief, the Word file in this folder, and assets/transcripts. Pick the passages for a [60-second] cut that supports the behavioral ask in my brief. Quote the words exactly as they appear, give each passage’s file name and its start and end timestamps, and list them in order as a table in paper_edit.md. Do not paraphrase or join sentences that were not spoken together.

Step C3. Script.

Script table, Track C

Read my communication brief, the Word file in this folder, and paper_edit.md. Write a script for a video of about [3 minutes] as a table in script.md, with one row per line or clip. Columns: number, type (narration or clip), the words, what is on screen, the speaker’s name and title, and the source. Clips use the exact words from paper_edit.md. It ends on one next step: [sign up / volunteer / call], with [the URL or phone number]. Use only facts from my brief and the transcripts. Where a line needs a fact I have not given you, write NEEDS SOURCE instead of filling it in.

Send script.md to the organization and correct every program detail, date and number before you build.

Step C4. Other material. Skip the parts your video does not need.

Clip cuts

Cut each passage in paper_edit.md from its original file with ffmpeg and save it in public/clips, numbered by script row. Add half a second of room at each end. Transcribe each cut with Whisper and tell me if any cut starts or ends in the middle of a word.

Screen recording, for a website or tool

Record [URL] at 1920 by 1080 with Playwright, following these steps at a pace a viewer can follow: [open the home page, type a search, click a result]. Save one MP4 per step in public/screens. Do not record any page with personal information. Show me a still from each.

Asset research, Track C

For rows of script.md that still need a picture, find public-domain or openly licensed material from Wikimedia Commons, Library of Congress and National Archives. Save each item in assets/photos. Record the license exactly as the source page states it and never guess one. Write assets/MANIFEST.md with the file, the source URL, the license and the credit line. Tell me which rows have nothing usable. If a site blocks a download, give me the URL and I will save the file myself.

For drawn figures, use the illustration prompt in Track A and label them as AI-made in the video.

Step C5. Narration. Use a staff member’s recorded voice with their permission, your own voice, or ElevenLabs. Clip rows keep the speaker’s own sound.

Narration, Track C

Read my ElevenLabs key from the environment variable ELEVENLABS_API_KEY; do not print it. Make one narration file for each narration row of script.md with the voice [name], and save them as public/vo/01.mp3 and so on, numbered by script row. Spell numbers out in the text sent to the voice. Then transcribe each file with Whisper on this laptop, and list any number or name that was read differently from the script, and the length of each file.

Step C6. Visual system.

Visual system, Track C

Read references/style.md and follow it where it does not conflict with the organization’s look. Set up the Remotion project using assets/org: its logo, its main color as the one accent, the font [Public Sans] for headlines and body text, and [Newsreader] for quotations. Every scene has a plain background. People on camera have a name and title at the lower left when they first appear. Quotations appear on a simple card with the speaker’s name. Make one test scene and show me a still before you build the others.

Step C7. Scenes. Send the matching prompt for one scene, check it, and repeat for the others.

Interview scene

Build scene [2] from row [2] of script.md using the clip in public/clips. Show it full frame with its own sound, and show the speaker’s name and title at the lower left for the first four seconds. Leave captions out of the picture; they go up as a separate file. The scene lasts as long as the clip.

Screen scene

Build scene [4] from row [4] of script.md using the recording in public/screens. Show it inside a browser frame, zoom slowly to [the search box] when the narration says [search], and draw a short numbered label beside it. The scene lasts as long as its narration file plus two-thirds of a second.

Next-step scene

Build the last scene from the last row of script.md. Show one sentence with the next step, the [URL or phone number] in large type, and the organization’s logo, on screen for at least six seconds. The narration reads the URL or number aloud.

Remotion Studio

Start Remotion Studio and tell me the address to open in my browser.

Look at each scene in the browser and describe what to change in plain words. Then go to Finishing.

Finishing, all tracks

Step 1. Render.

Render

Join all scenes in order with their narration and render the MP4 at 1.5 times the composition size, so it comes out 1920 by 1080, to out/video.mp4. Then take one still from the middle of every scene and put them on one sheet so I can check them side by side. Tell me the total length and the resolution of every footage file you used.

A soft render usually comes from a small footage file. Ask for the largest version of that item. The assignment asks for 3 to 5 minutes.

Step 2. Sound.

Sound

Find a quiet background music track and any small sound effects the script calls for. Use only recordings with a license that allows use on YouTube (the YouTube Audio Library, Pixabay, or Creative Commons music), record each license in assets/sound/MANIFEST.md, and give me the credit lines for the description. If a site blocks a download, give me the URL and I will save the file myself. Mix them under the narration about 15 dB lower than the voice, lower still while someone speaks, without rendering the video again.

The agent cannot hear the result. Listen to the whole video.

Step 3. Captions.

Caption file

Write captions.en.srt for out/video.mp4. Use the words of script.md, not a machine transcript, and use Whisper on this laptop only to time them. Each cue has at most two lines of 42 characters. Include any speech in clips and label other languages in brackets. Tell me if any cue is faster than about 20 characters per second.

Watch the video with the captions on and correct names, numbers and non-English words by hand. The assignment asks for this corrected file.

Step 4. Accessibility audit.

Accessibility audit

Read script.md and captions.en.srt, and check one still from every scene. Audit the video for accessibility and give me a table with the time, the problem and the fix. Check: caption lines over 42 characters or cues over two lines; cues faster than about 20 characters per second; speech, other languages and sounds missing from the captions; on-screen text the narration never reads aloud (sources, labels, quotes, highlighted lines), listed by scene; text smaller than 24 pixels at 1080p or low in contrast against its background; anything that flashes more than three times a second.

Step 5. Transcript.

Descriptive transcript

Write a descriptive transcript of the video as plain text in transcript.txt: the narration and any other speech in order with a timestamp, and after each scene, in brackets, what the screen shows that no one says aloud, such as photo captions, labels, quoted statements, highlighted lines from documents and source lines. Mark speech in another language and give its English translation with the source of that translation.

Step 6. Captions in other languages. Optional.

Caption translation

Translate captions.en.srt into [language] and save it as captions.[language code].srt. Keep every index number and timestamp line exactly as it is, change only the text lines, keep at most two lines per caption, and use the terms US agencies use in that language for legal and program names. Check the structure against the English file with a short script and report any difference.

Then open a new chat in Claude, Gemini or UM-GPT and ask it to review the translation line by line against the English, listing every change with the caption number, the old text, the new text and the reason. Label the tracks “AI translations” in the description until someone who reads the language has checked them.

Step 7. Cover and description.

Cover

Make a 1280 by 720 cover image in the same visual system as the video, with the title in the headline font and one image from the video. No politician’s face and no generated person. Save it as out/cover.png.

Video description

Write a video description in description.txt: two sentences on the video, chapters with times, the link to the transcript, and every credit line from the MANIFEST files. List which images and voices are AI-made. Use only facts from script.md and the MANIFEST files.

Step 8. Final checks. Go through the checks before publication. List your AI prompts in the reflection, as the assignment asks. For Track C, send the cut to the organization before it screens.

Common problems

Common problems, with the next step for each
ProblemNext step
agy is not recognizedClose the terminal and open a new one. On Windows, the program is in AppData\Local\agy\bin
An error message appearsCopy all of it into the chat and add: “Explain this in plain language, then show me how to fix it.”
The agent goes in the wrong directionPress Ctrl+C and say what you want instead
A scene came out wrongDescribe what you see, one change at a time, and name the scene number
The agent says the quota is used upWait for the refresh, or type /model to choose another model
The session closedOpen a terminal, type cd my-video, then agy -c

Reusable skills

A skill is a saved set of instructions you can call by name, in any later video. Antigravity reads skills from a project’s .agents/skills folder (Antigravity skills documentation).

Save a skill

Create a skill at .agents/skills/my-video-style/SKILL.md, based on the visual system and scene rules in this project. Start the file with name and description lines in YAML frontmatter, and write this description: “Use when building a new scene for my video.” List only rules we actually used here.

Type /exit, reopen with agy -c, and call it by typing /my-video-style.