drone
Aerial Reveal: Pull-Back Over Lone Surfer at Golden Hour
Use specific camera-motion verbs ('pull-back', 'rotate revealing') instead of generic 'drone shot' — recommended pattern from community prompt collections.
Prompt
Aerial drone shot beginning tight on a lone surfer paddling on glassy water at golden hour. Slow upward pull-back and rotate revealing 800-foot Big Sur sea cliffs, ribbons of mist clinging to the rock face. Warm golden light from low west, deep teal water below. 10-second continuous shot, 16:9, smooth cinematic motion (no jitter).
Verified output — the hardest camera move we have landed cleanly
The video above is the unedited Omni Flash output from this exact prompt (generated 2026-05-29 on Google Flow, 10s 16:9 720p, SynthID-watermarked). The pull-back-and-rotate reveal is the most complex camera move in our library, and Omni held it in one continuous shot with no cut and no jitter:
- The reveal arc landed frame-for-frame — the clip opens tight on the surfer paddling on glassy, golden-lit water, then the camera rises and rotates to expose the full Big Sur cliff line with mist ribbons clinging to the rock face. Exactly the progression the prompt describes.
- The “smooth cinematic motion (no jitter)” cue worked — none of the nervous micro-shake that aerial prompts often produce. The motion reads like a stabilized drone.
- Pulling back, not pushing in, kept the human clean — our failure-mode notes warn that “people becoming central as the camera approaches is unreliable.” Here the surfer starts central and shrinks into a scale reference, so Omni never has to resolve a face or hands up close. That is the safe direction for a human in an aerial shot.
- Color brief nailed — warm low-west golden light against deep teal water, exactly as written. Naming two contrasting elements (warm light / cool water) is the most reliable cinematic lever in Omni.
Takeaway: complex camera moves do work in Omni Flash if you (1) name a specific motion combo (“upward pull-back and rotate revealing X”), (2) add an anti-jitter cue, and (3) move the camera away from any human rather than toward them.
Why this prompt
Generic “drone shot” doesn’t give Omni enough information — the model has dozens of possible drone-style interpretations. The working pattern (per community prompt collections):
- Open with subject and starting frame
- Use specific camera-motion verbs:
pull-back / ascend / orbit / sweep / push-in / dolly - Anchor with at least one environment detail: mist / fog / surface texture / weather
This Big Sur reveal appears in geminiomnivideo.io’s curated prompts across multiple aerial categories.
Source tier: 🟢 Recommendation tier — pattern documented across multiple sites, not independently verified with public output
Camera motion vocabulary that Omni recognizes
From DeepMind’s prompt guide, these terms are explicitly parsed:
- Vertical motion: pull-back / pull-up / pull-down / ascend / descend
- Horizontal motion: sweep / pan / dolly / orbit
- Push motion: push in / punch in / dolly zoom
- Stable: locked off / fixed / oner / continuous shot
Combine motion verbs: “slow upward pull-back and rotate revealing X” is more reliable than “the camera moves up and around”.
How to tweak
- Location: Big Sur cliffs → tropical island lagoon / desert canyon / Icelandic glacier / Norwegian fjord
- Subject anchor: surfer paddling → kayak / fisherman casting / lone hiker / waterskier
- Time of day: golden hour → blue hour / midday hard shadows / overcast moody
- Atmosphere: mist on cliffs → fog over lake / snow flurries / sea spray / rainbow after storm
- Aspect ratio: 16:9 hero shot → 2.39:1 cinematic widescreen → 9:16 Reels (use 6-7s)
Common failure modes
- “DJI Mavic” or specific drone brand names: Omni doesn’t parse drone hardware brands. Use motion verbs instead.
- Multi-stage reveals over 10s: Flash limit. For complex reveals, stick to one motion phase per clip.
- Drones + people interaction: Omni handles wide aerial well, but people becoming central as the camera approaches is unreliable.
- No anchor detail: “drone shot of mountains” → generic. “Drone shot revealing fog-shrouded peaks at dawn” → specific output.
Notes
- 10 seconds is the Flash hard limit (TechCrunch confirmed)
- “Smooth cinematic motion (no jitter)” hint prevents over-shaky output
- For YouTube hero footage, this 10s clip can become 60s by chaining 6 variations (different angles, times of day) via conversational editing
Sources
- Pattern documentation: geminiomnivideo.io best prompts
- Camera vocab: DeepMind prompt guide
- Duration limit: TechCrunch launch coverage
Related