|Docs

Cut an event sizzle reel and a town hall recap from one day of footage

Turn a day of conference, launch event or all-hands footage into two videos: an event sizzle reel, or highlight reel, with music and titles, and a one-minute town hall recap of the speakers' best lines with captions. You also get an SRT captions file for your intranet player. Your agent finds loud crowd moments for you to check, and the strongest quotes, so you do not scrub through hours of footage.

Use invideo for this when you need both cuts by the next morning from the same footage, each on its own branch you can review and change in the editor.

The agent never posts to your intranet, your social accounts or Slack. You publish.

Quickest path

Set up your agent first with invideo for AI agents. Then give it this prompt.

Here is yesterday's event footage in ./event. Make two cuts on twobranches: a 60-second sizzle reel cut to upbeat music with the eventname as a title, and a one-minute recap of the CEO's and the panel'sbest lines with captions. Give me an SRT for the recap. Price thetranscripts and music first.

In a chat app, you drag the files into the project. In a terminal, the agent names the folder and uploads it.

What you need

These are the inputs the agent asks for.

  • The footage: stage cameras, the room camera, phone clips, the stream recording.
  • The names and titles of the speakers, spelled right.
  • The event name, date and logo, and your brand colors.
  • A music mood for the sizzle reel, or a track you own.
  • Who must be in the recap, and anything that must stay out, like a confidential slide.

Steps

The agent finds the moments first, then cuts the two videos from them.

Find the moments

The agent builds one selects timeline, then branches it into the two cuts.

  1. Upload the footage

    Event files are large, so the agent uploads each with start_upload, PUTs the bytes, then calls add_media with the upload_id. Expected result: one sequence URL per file.

  2. Find the loud moments and the scene changes

    The agent runs media audio.onsets and audio.content on each file for loud hits, sound effect spans and music. These are candidates for applause, cheers and laughs, not labels. It runs video.sceneSplit on long camera files to find the cuts, and video.frames at each candidate to look. Expected result: a list of moments with times and a frame each.

  3. Find the best lines

    The agent prices audio.transcript on the sequence URL of each talk and the panel with dry_run, tells you the price, starts it with max_credits set to that price, and calls again after retry_after_s for the words. It picks the short, complete lines that land without context. Expected result: a list of quotes with the speaker, the file and the times, shown to you.

  4. Build the selects and branch

    The agent makes a 16:9 timeline with new_timeline, places the picks with place_media, and marks each with add_marker. Then it runs timeline_history with create_branch twice, named sizzle and recap. Expected result: two branch URLs from one selects timeline.

Cut both videos

The agent cuts each video on its own branch, then checks both before export.

  1. Make the music

    The agent runs find_models with the job in words, picks one, and reads its inputs with get_model. A good pick is elevenlabs_music_v2_5_instrumental. It calls generate with dry_run: true, a prompt and a duration, and tells you the price. Then it calls generate with max_credits set to that price and an idempotency_key. Expected result: a sequence URL, which the agent reads with read.

  2. Cut the sizzle reel

    On the sizzle branch, the agent runs audio.beats on the track for the beats and audio.markers for the drops and phrases. It cuts on the beats with split_clip, trim_clip and delete_clip, puts the biggest moment on a drop, and adds the event title and the logo with add_text and place_media. Expected result: a tight cut that ends on the music's last beat.

  3. Cut the recap and caption it

    On the recap branch, the agent keeps the chosen lines in order, adds a lower third per speaker with add_text, and adds captions with add_captions from the transcript words. Expected result: a cut with captions that match the speech.

  4. Check, export, and write the SRT

    The agent runs capture at each cut and check_timeline on both branches, then export each branch with format video. For the SRT, it reads the recap's caption slides with get_timeline mode collection and collection caption_slide, and writes the file. Expected result: two export cards with status completed, and an SRT file for the recap.

The terminal version of the same job:

invideo project new --title "Summit recap"invideo agent register <project-url> --name <name>export INVIDEO_AS=<name>invideo asset upload <project-url> ./event/*.mp4invideo media audio.onsets <sequence-url>invideo media audio.content <sequence-url>invideo media video.sceneSplit <sequence-url> --dry-runinvideo media audio.transcript <sequence-url> --dry-runinvideo media audio.transcript <sequence-url> --max-credits <price>invideo edit new <project-url> --name "Summit selects"invideo edit show <edit-url> --jsoninvideo branch new <edit-url> --name sizzle --token <state_token>invideo edit show <edit-url> --jsoninvideo branch new <edit-url> --name recap --token <state_token>invideo edit show <sizzle-url> --jsoninvideo gen models <project-url> --query "upbeat instrumental music"invideo gen model <project-url> --model elevenlabs_music_v2_5_instrumentalinvideo gen run <page-url> --model elevenlabs_music_v2_5_instrumental \  --dry-run -i prompt="driving, bright, festival energy" -i duration=60invideo gen run <page-url> --model elevenlabs_music_v2_5_instrumental \  --max-credits <price> --idempotency-key summit-music-v1 \  -i prompt="driving, bright, festival energy" -i duration=60invideo media audio.beats <music-seq-url>invideo media audio.markers <music-seq-url>invideo edit change <sizzle-url> --timeline T1 --token <state_token> \  --operations @sizzle.jsoninvideo edit show <recap-url> --jsoninvideo edit change <recap-url> --timeline T1 --token <state_token> \  --operations @recap.jsoninvideo edit show <recap-url> --mode collection \  --collection caption_slide --json > slides.jsoninvideo edit check <sizzle-url>invideo edit export <sizzle-url> --format video --quality 1080 --jsoninvideo status <card-url> --waitinvideo notebook show <card-url> --out ./sizzleinvideo edit export <recap-url> --format video --quality 1080 --json

Where the values come from:

  • <project-url> comes from invideo project new, and <edit-url> from invideo edit new. <name> is the name you choose in invideo agent register. <page-url> is a notebook page from invideo notebook show <project-url>.
  • <sequence-url> is the url the upload returns for each file. Each <price> is the dry run right before it.
  • <music-seq-url> is the sequence URL from the music run or the stock pick.
  • <state_token> comes from invideo edit show <edit-url>, read fresh before each branch new.
  • <sizzle-url> and <recap-url> are the .branch_url each invideo branch new returns. Before each change, <state_token> is the one invideo edit show returns on that branch; the main token is stale there.
  • <card-url> is .sequences[0] of what each export returns. invideo status waits and does not download; invideo notebook show saves the file as ./sizzle/export.mp4. The recap export gets its own status and notebook show.

The file ./recap.json holds a bare list of operations for the recap branch. On the MCP path, the same list goes in the operations field of edit_timeline. It adds a speaker's lower third and the caption for one line, both on V3 (Titles). The lower third takes that span first, so add_captions puts the captions on the next free video track above it. At 30 fps, frame 90 is 3 seconds.

[  { "op": "add_text", "track": "V3", "start_f": 0, "end_f": 90,    "text": "Maya Sen, CEO",    "style": { "font_size": 42, "color": "#ffffff",      "align": "left", "shadow": true } },  { "op": "add_captions", "track": "V3",    "slides": [      { "text": "This year the team shipped more than in the last three.",        "start_f": 0, "end_f": 96 }    ] }]

The SRT file

The agent turns each caption slide into one SRT block: a number, the start and end time as hours, minutes, seconds and milliseconds, then the text. It converts frames to time at the timeline's rate. Then it gives you the file to upload with the video.

100:00:00,000 --> 00:00:03,200This year the team shipped more than in the last three.

Cost and time

Edits and exports are free. Generation and transcripts cost credits. Paid stock costs a small credit charge, or one of your workspace's iStock licenses. Your agent runs a dry run first and tells you the price before it spends anything.

  • Free: every edit, the SRT, capture and export.
  • Costs credits: audio.transcript on each talk, stock music or a generated track.
  • The media reads (audio.onsets, audio.content, audio.beats, audio.markers, video.frames, video.sceneSplit): the dry run shows the price before anything starts.

The upload and the transcripts take the longest. Plan the evening after the event for both, and the cuts the next morning. See Credits for prices.

If something goes wrong

Most problems come from long files or loud rooms.

SymptomFix
get_model or generate refuses the model as not foundModels retire. Run find_models with the job in words, use the best match, and tell the person which model you used.
Exit code 6, or status is heldA safety review, not a refusal. Wait, then run invideo status <url> --wait again. Do not resend. If it ends failed, change the input the review flagged.
The upload PUT failsCall start_upload again for a new upload_url, and PUT once more.
A file has no sound, so it has no audio cuesPick its moments from video.frames and the scene cuts.
You want stock music instead of a generated trackRun search_library in stock_music, and bring the pick in with add_media.
The applause list is mostly crowd noiseCheck each candidate with video.frames. Keep the ones where the stage shows a moment.
A quote starts mid-wordMove the in point to the gap before the first word from the transcript.
The SRT drifts from the speechThe frame-to-time math used the wrong rate. Read the timeline's rate with get_timeline and write it again.
audio.transcript returns a status, not wordsIt is still running. Call again with the same file after retry_after_s.
An edit on one branch shows up on the otherIt does not. Check that the URL names the right branch.
CLI exit 3The edit changed since your read, or the token is from main. Run invideo edit show on that branch's URL and use its state_token.
The ./sizzle folder is empty after statusstatus waits and does not download. Run invideo notebook show <card-url> --out ./sizzle.
CLI exit 7Nothing ran. Run the same command again.
Avoid: one timeline for both cutsInstead: one branch per deliverable, so a change to one never breaks the other.
Avoid: keeping a confidential slide in frameInstead: check each frame with capture, and cut or blur what the person named.

Prompts to try

Each prompt is a real variation of this job.

Make the sizzle reel in 9:16 too, on a third branch, with the title atthe top and no text near the bottom.
The town hall has no slides worth showing. Cut the CEO's five best linesinto one minute with captions and the CEO's name as a lower third.
Make a 15-second teaser from the sizzle branch for tomorrow's newsletter,ending on the logo.

When to use something else

invideo does not record or stream the event, post to your intranet, or make SCC caption files. Use your streaming tool to record, your intranet or video platform to publish, and a captioning tool such as Rev or 3Play Media if your platform needs SCC. For a multi-camera live switch during the event, use a production switcher.