POV hiking video prompt for Gemini Omni: handheld, breath, steps

A POV hike works in Gemini Omni when the prompt says whose eyes the camera is, how it moves, and what you hear from your own body. Write 'first-person, handheld, chest height', add boots, breath and wind, and ask for one continuous shot. A 10-second vertical clip is $1.25 at 720p on Sume.

POV is a forgiving genre because the subject, the walker, is mostly off camera; hands and boots are the only body parts to get right.

Camera words that read as POV

Omni understands informal camera language; Google's guide says to describe the shot in plain terms (read 2026-10-07). Use these.

• 'First-person view, handheld at chest height, slight natural sway.'
• 'Hands with trekking poles enter the bottom of frame.'
• 'The trail winds uphill toward a ridge; camera tilts up at the end.'
• 'Single continuous shot, no cuts.'

Sound from the body

Name three sounds only: footsteps on gravel, steady breathing, and wind. A fourth, such as birds, is fine; a fifth muddies the mix. End with 'no music, no dialogue' unless you want a score. The audio comes from the same model as the picture and cannot be switched off with a parameter on Sume, so the prompt is the control.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: pov-trail-1" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "First-person view, handheld at chest height with slight natural sway. Gloved hands hold trekking poles at the bottom of frame as the walker climbs a gravel trail toward a ridge at sunrise. Single continuous shot, no cuts. At the end the camera tilts up to the valley. Audio: boots on gravel, steady breathing, light wind. No music. No dialogue.",
    "duration": 10,
    "resolution": "720p",
    "aspect_ratio": "9:16"
  }'

Price

Sume bills the list rate times 1.25 per second, rounded up to the cent per job. fal's list for Omni Flash 1.1 (recorded 2026-08-28) is $0.03, $0.10, $0.15 and $0.30 per second at 360p, 720p, 1080p and 4K.

Stitching a longer hike

Omni stops at 10 seconds per job on Sume, and Sume does not expose Google's extend feature (the Video Router models page lists no extend mode). A longer hike is several jobs. To keep continuity, grab a still from the last second of the first clip with the video frames endpoint and send it as the first frame of the next job. Hold weather, light and clothing words the same across jobs, and keep the same aspect ratio so a timeline join needs no crop.

Planning the edit

A hike is rarely one shot. Plan three clips: a start at the trailhead, a climb, and a view at the ridge. Keep each at 10 seconds or less, generate them in parallel, and cut between them on a boot step or a camera tilt. Matching hand position at the cut (poles in the same hand) sells the join. If the weather word differs between clips ('overcast' in one, 'sunrise' in another) the edit will look like three days; write one light description and reuse it exactly.