Blog / 5 min read · 2026-05-24

Camera vocabulary Gemini Omni parses literally — verbs, lenses, and what to avoid

A working list of camera motion verbs and lens descriptors Gemini Omni interprets correctly. Plus the brand-name traps that look like camera language but actually do nothing.

Most Gemini Omni tutorials list “good prompt examples” without separating which words the model actually parses and which are decoration. This post is the working vocabulary: terms Omni reliably interprets, structured by what they do.

If you’re new to Omni, read the base formula in the field guide first. Then come back here for the verb list.

The six motion verb families

Omni’s prompt guide enumerates these six camera motion verb families. Each maps to a specific behavior you can rely on.

Push (forward motion toward subject)

• push in / slow push / fast push — dolly-in equivalent
• punch in — sharper, faster push
• dolly zoom — combined push-in + zoom-out for the Hitchcock effect

Use for: building intensity, isolating the subject, creating intimacy. Pair with a subject doing nothing — let the camera do the work.

Pull (reverse motion away from subject)

• pull-back / slow pull-back
• pull-up — pull-back + vertical rise (drone reveal)
• pull-down — pull-back + descend

Use for: revealing context, opening a scene, mood-shifting from intimate to expansive.

Reveal (motion that uncovers hidden information)

• ascend revealing the city below
• pull-back and rotate revealing X
• tilt up revealing the spire

Different from pure pull — the “revealing” destination is the syntactic anchor. Specify what’s being revealed for stronger results.

Orbit (camera revolves around subject)

• orbit around / slow orbit / 360 orbit
• sweep around
• circle the subject

Use for: hero product shots, character introductions, dramatic reveals. Combine with while subject stays still to lock subject behavior.

Pan (rotation from a fixed position)

• pan left to right / pan right to left
• vertical pan upward / vertical pan downward

Critical distinction from tracking: pan is rotation only, no spatial movement. Use for scanning a static scene.

Static (no camera movement)

• static shot / locked off / fixed camera
• oner — implies a single continuous take
• one continuous shot — explicit “no edits” instruction

Use when subject motion or staging carries the shot. The “one continuous shot” phrase is particularly load-bearing in Omni — it acts as a hard constraint, not a suggestion.

Style references Omni renders

These describe the medium the camera represents rather than the camera move:

• natural smartphone zoom — slightly soft, digital
• film camera, warm, slight grain — film-stock texture
• 16mm film — coarser grain, vintage palette
• 35mm film — cleaner cinematic look
• webcam style — compressed, slightly soft (good for video-call mockups)
• handheld, micro-jitter, organic — humanized motion
• stabilized handheld — minimal shake
• shaky handheld — chaotic, found-footage

Pair one style reference with one motion verb. Stacking many (“16mm film, handheld, dolly zoom, anamorphic flare”) dilutes focus.

Composition + duration locks

Lock these in your opening sentence:

Create a [duration]-second [aspect-ratio] [genre] video in one continuous shot.

• 8-second 16:9 — standard cinematic
• 5-second 9:16 — Reels / Shorts / TikTok
• 10-second 1:1 — square social (max Omni duration)
• 8-second 2.39:1 — anamorphic widescreen

Omni respects these strictly when specified up front. Stuffing them in metadata or at the end of the prompt is less reliable.

What NOT to write

These look like camera language but actually do nothing in Omni:

Camera brand names

• shot on DJI Mavic Pro ❌
• RED Komodo footage ❌
• BMPCC 6K Pro ❌

Omni doesn’t parse hardware brands. Use motion verbs and style references instead. “Drone-style aerial pull-up over the cliff” beats “shot on DJI Mavic 3 Cine” every time.

Adjective stacks

• cinematic dramatic moody atmospheric epic stunning beautiful breathtaking ❌

DeepMind’s prompt guide explicitly says Omni “doesn’t need to be as prescriptive as Veo” — natural language with one clear cinematic anchor beats adjective lists.

IP-bound style names

• Studio Ghibli style ⚠️ — works but raises IP flags
• Pixar 3D animation style ⚠️ — same

Use medium descriptors instead: hand-painted watercolor animation, 3D animated, soft rim lighting, warm palette. Same look, no legal exposure.

Words over 50

Per Seaart’s analysis, prompts longer than ~50 words dilute focus. Be specific but concise. If your prompt grows, split into a base scene + a follow-up turn with the trigger pattern.

Quick reference: a working prompt

Create an 8-second 16:9 cinematic video in one continuous shot.
A woman in a red coat walks through a snowy Tokyo alley at night.
Slow handheld tracking from her side. 35mm film, warm tungsten lighting from windows.

Parses to: 8 sec duration ✓, 16:9 aspect ✓, one continuous shot ✓, subject ✓, action ✓, location ✓, time of day ✓, camera motion verb (tracking) ✓, style reference (35mm film) ✓, lighting ✓. Under 50 words. No brand names. No adjective stacks.

That’s the shape Omni produces reliably.

Related

• Full field guide — base formula and opening-line trick
• “Keep X identical” lock — for follow-up turns
• Trigger pattern for VFX — when you need a transformation mid-clip
• Glossary — full term-by-term reference