Gemini Omni 1.2 Flash: 15-Second AI Video Generation & 5 Creative Workflows

15 Seconds: Enough for a Mini Daily Scene

1. Selfie Street Walk: Ready-Made Shop Exploration Opener

In Ozan Sihay's demo, a boy holds a selfie camera walking down a street. He passes a shop clerk, an elder on a bench, and a few pedestrians, finally looking back at the lens. The protagonist keeps walking while others appear along his path, creating a mini narrative in 15 seconds.

Selfie perspective strings together several character interactions; source: Ozan Sihay

Swap in a coffee shop: enter from the door, greet the owner, receive a coffee, and end. A shop-exploration opener is ready.

2. Anime Cooking: A Few Details Deliver Flavor

This clip kneads dough, mixes filling, shows character close-ups, and steams dumplings — several shots assembled into a short anime daily-life piece.

Everyday details compose an anime short; source: Gorden Sun

Brewing tea, toasting bread, organizing a desk — all follow the same pattern. For tea, the prompt might be: "Warm anime style, shoot the hand pouring tea, the person waiting, finally return to steam rising from the cup." Character actions and camera shots are packed into one sentence.

3. Let the Drawn Character Run into a Real Room

Higgsfield's demo shows a cartoon fridge magnet peel off, become a 3D character, and share the frame with a real dog in a living room. Flat art and live action merge like a character entrance trailer.

Flat character enters a realistic living room; source: Higgsfield platform demo

Hand-drawn characters, figurines, brand mascots can all go this direction. Feed the model a character image and a room photo, then direct where it appears, walks, and interacts. Reference images tell the model: this face, this outfit, this product must persist in later shots. Google says the new version tracks these details better. In such outputs, I watch faces, patterns, and package text closely.

Keep the Motion, Swap the Subject

4. Octopus Motion Transferred to a Dancer's Skirt

Google gave an intuitive demo: an octopus extends its body; swapped to a dancer, the skirt inherits that soft, flowing feel. The octopus tentacles' unfurling becomes the skirt's rise and fall.

Retain flowing motion feel, then replace subject; Google official demo, shared by Min Choi

Multi-turn editing also works for precise tweaks. A product rotates well; keep the motion, change the table, change the background — just add the request in the same conversation.

Example prompt:

💡 PROMPT
Prompt:
Only change the background to a wooden tabletop. Keep the product's shape, color, packaging text, rotation speed, and camera position unchanged.

Each turn changes one thing, making before/after differences easy to spot. Keep satisfactory parts, iteratively shape the clip to the desired result.

5. Educational Long-Form: Have AI Split It into Segments

Chetaslua showed a ~115-second jet engine explainer. Shots move from exterior to internal structure, using flow lines and combustion visuals to explain operation.

From overall view to internal structure explainer shots; source: Chetaslua

The author used Gemini 4 Carbon to plan the piece, then called Omni Flash 1.2 to generate frames. Product internals, natural phenomena — both can follow this workflow: write narration, break into shots, generate segment by segment.

Omni's single-segment limit is 15 seconds; this 115-second video relies on compositional production.

Fight Scenes Test Motion Fidelity Immediately

"风雪漫千山" posted a xianxia fight comparison: top half Seedance 2.5, bottom half Omni Flash 1.2. Sprint, clash, VFX release — side-by-side reveals clear differences in motion and cut continuity.

Top: Seedance 2.5, Bottom: Omni Flash 1.2; source: 风雪漫千山

The original author shared a detailed long prompt: character attire, seconds per shot, camera movement — all specified. Even with such detail, it comes back to motion itself: where fists land, how bodies react to hits, whether positions match after cuts. For animated drama, these matter more than VFX flashiness.

Fast motion and complex physics are areas Google notes are still improving. Simple walking, product showcase shots vs. full high-speed combat — the difficulty gap is real.

Access Points and Costs

Entry points: Google Vids, Gemini App, or Flow's video feature — use whichever is convenient.

Gemini App : New version rolling out to Google AI subscribers; 15-sec generation and editing limited to Ultra tier

Google Vids : Announcement states 15-sec clips at no extra fee; quota depends on Google or Workspace account limits

Flow & Third-Party Tools : Subject to each platform's plans, credits, and regional rules

Rollout is gradual; feature appearance timing varies. The entry points and quotas visible in your account are what you can actually use.

Pricing: Google cut the 720p developer API fee by 30% vs. previous v1.1. Now approximately $0.07 per second, ~$1.04 for a 15-second video output . App and third-party tools charge per their own plans.

On API billing, drafts, revisions, and HD outputs accumulate costs; input assets also incur fees. I'd test motion and composition at 360p or 720p, then pay for 1080p or 4K upscale — the latter two are upscale outputs billed separately.

Draft, revision, and HD output costs accumulate