Gym promo for Reels: three beats in 8 seconds with Gemini Omni

A vertical gym promo fits in one 8-second Gemini Omni clip if you write three timecoded beats: arrive, effort, payoff. Google's Omni guide shows the syntax "[0-3s] A person is walking [3-6s] They stop and turn around [6-10s] They start running" (read 2026-10-07), and Sume accepts that text in the prompt of POST /v1/videos with aspect_ratio: 9:16. At 720p the clip costs $1.00.

Keep the brief to one person doing one ordinary lift, so each beat has a single clear action.

Beat sheet for 8 seconds

Three beats keep the pace readable on a phone. Give each its own range and one verb.

• [0-3s] Arrive: 'a trainer chalks her hands and steps up to a barbell, low angle'.
• [3-6s] Effort: 'the lift, camera orbits left, plates rattle, breath visible'.
• [6-8s] Payoff: 'she sets the bar down, exhales, small smile, camera settles'.
• Audio line: 'gym ambience, bar clang, one heavy exhale. Upbeat electronic loop at low volume. No dialogue.'

The request

Everything is in the prompt; nothing else needs setting except the vertical ratio. Sume rejects ratios Omni does not offer (only 16:9 and 9:16 are accepted), so a 4:5 feed crop happens later.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: gym-promo-1" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Vertical gym promo, single continuous shot. [0-3s] a trainer chalks her hands and steps up to a barbell, low angle. [3-6s] she lifts, camera orbits left, plates rattle. [6-8s] she sets the bar down, exhales, small smile. Audio: gym ambience, bar clang, low upbeat electronic loop. No dialogue.",
    "duration": 8,
    "resolution": "720p",
    "aspect_ratio": "9:16"
  }'

Price and draft ladder

List rates in the Video Router doc are $0.03, $0.10, $0.15 and $0.30 per second at 360p, 720p, 1080p and 4K, and Sume bills list times 1.25, rounded up to the cent per job.

What to verify before posting

Watch the middle beat: if the lift is the part that looks wrong, shorten it, because lifting form is something viewers who train will notice. If a beat is skipped, the usual fix is fewer words in that range. Remember that the native audio is always on and cannot be turned off by a parameter on this model, so check that the soundtrack is something you may publish before you upload it to a platform with its own music rules.

Platform checks before posting

Short vertical ads are reviewed by the platforms that carry them, each with its own rules for synthetic media and music. Sume's docs do not describe those rules, so read the platform's current policy for the place you will post. Google says every Omni video carries an invisible SynthID watermark (read 2026-10-07), which is a provenance mark, not a label that viewers see. For a gym or a studio, avoid showing identifiable real people or real brands in the generated footage, and keep health claims out of the prompt and the caption.