White elephant invite video: 10-second AI clip, rules on screen

To make an office white elephant (or Secret Santa) invite video, generate one short clip of wrapped gifts on a table with gemini-omni-flash-1.1, which takes 3 to 10 seconds with native synced audio, then burn the event details into it as authored caption cues: date, time, spend limit and the steal rule. The clip sets the mood; the cues carry the facts, so nobody has to trust an AI model to spell a time correctly.

Both calls are documented on Sume's API pages: POST /v1/videos for the clip and POST /v1/video-captions for the text.

Which model fits a 10-second invite?

Sume's video docs list Gemini Omni Flash 1.1 as text, image and reference-to-video for 3 to 10 seconds with native audio. An invite needs about that length: long enough for a gift-table pan, short enough to play in a team chat. A text-only request works; if you have a photo of your office or a real gift stack, pass it as a reference image instead.

How do you put the rules on screen?

Do not ask the video model to render the date. Instead, send the finished clip URL to the caption endpoint with cues, which the captions page describes as authored overlay text that needs no speech-to-text. Each cue has text, start and end. If your clip has no spoken words, plain speech captions fail with caption_no_speech, so cues are the right path.

Good cue lines for an invite: the date and room, the spend limit, and one rule line such as how many times a gift can be stolen. Keep each line short; the docs let you tune phrasing with max_words and max_chars under design.

• Cue 1 (0-3 s): the day, time and room
• Cue 2 (3-6 s): the spend limit
• Cue 3 (6-10 s): the steal rule and an RSVP deadline

How do you run it without babysitting?

Video jobs are asynchronous. Send a HTTPS callback_url on the video request and let Sume tell you when it is done, or poll the job roughly every 30 seconds, as the video page recommends. Then pass the finished file URL to the caption call. If the job fails, you can see the status and error on the job envelope described in the jobs page rather than guessing.

{
  "video_url": "https://example.com/invite.mp4",
  "style": "black-outline",
  "cues": [
    {"text": "Fri Dec 18, 3 pm, Room 4", "start": 0, "end": 3},
    {"text": "Gifts $25 max, wrapped", "start": 3, "end": 6},
    {"text": "Steal twice, then it is locked", "start": 6, "end": 9}
  ]
}

What should you check before sending?

Watch it once with the sound off, since many people open chat links muted. Confirm every cue is readable against the footage. If the plan changes, re-run only the caption call on the same clip rather than regenerating the video. The date shown above is an example; use your own.