Gemini Omni 1.2 Flash Text to Video Usage Guide
Write the prompt
Describe the scene, the motion and the sound in one prompt. Gemini Omni 1.2 Flash follows a sequence well, so give beats in order ("first..., then..., finally...") or use a timecode such as [0-3s]. Name a stylised medium, for example flat 2D motion graphics, claymation, paper collage, low-poly or 3D, rather than asking for photorealism. That is where the model is strongest and where it avoids content-safety blocks.
Put text on screen
To render words, quote each one exactly in single quotes and end the prompt with "every word spelled exactly as given, no other text". Gemini Omni 1.2 Flash renders titles, labels and numbers legibly, which suits infographics, explainers and product callouts.
Fix the layout and camera
For data videos and fixed layouts, say "locked-off camera" and list what must not move, such as the background, title bar or grid. Only the elements you animate will move.
Design the sound
Write the sound design beat by beat, one sound per event, for example a whoosh as a title slides in, a note as each bar grows and a chime on the final tag. Audio is generated with the video in a single pass.
Recommended Settings
- •duration: 6 seconds suits a short multi-beat explainer; use 10-15 for longer sequences. Range is 3-15.
- •resolution: draft at 360p for speed, then regenerate the winning take at 720p or higher. On-screen text needs 720p or above to stay legible; deliver finals at 1080p or 4K.
- •aspect_ratio: 16:9 for landscape, 9:16 for vertical and social.
- •seed: leave at -1 for variety; set a fixed integer to reproduce a take.
The tested default (6 seconds, 720p, 16:9) produced a clean flat-vector bar-chart infographic with its title, three labelled bars and tag all legible and in order, plus a matched soundtrack. Expect about one instruction in four to be skipped in dense prompts, so check the frames and text before upscaling.
Examples to try
Each example ran end to end at the settings shown, with the picture and native audio made in a single pass; click a card to load its prompt and settings into the form.
- •Flat-vector data infographic with synced audio (the default example). Prompt: "Flat 2D vector motion-graphics infographic, locked-off camera, deep teal background, bold white sans-serif type. First a title bar slides in from the left reading 'CITY RAINFALL'. Then three rounded blue bars grow upward one after another, each labelled beneath it 'JAN', then 'MAY', then 'SEP'. Finally a yellow tag pops in at the top reading 'MM PER MONTH'. The background grid and the title bar stay still; only the bars and the tag animate. No people, no photorealism. Sound: a soft paper whoosh as the title slides in, a short rising marimba note as each bar grows, a light chime when the tag appears. Every word spelled exactly as given, no other text." At 6s/720p/16:9 the title, three labelled bars and the tag arrive in order, each with its own sound. Tip: quote every on-screen word in single quotes.
- •A stylised medium beyond flat vector — stop-motion claymation. Prompt: "Stop-motion claymation scene on a tiny clay kitchen table, single continuous shot, locked-off camera. First a round clay teapot tips forward and pours a ribbon of amber clay tea into a white clay cup. Then a clay sugar cube drops in and the surface ripples. Finally a clay lemon wedge tips onto the saucer. The table and background stay still; only the teapot, tea, sugar and lemon move. No people, no text, no photorealism. Sound: a soft gurgle as the tea pours, a gentle plop as the sugar drops, a tiny clink as the lemon lands. No dialogue." At 5s/720p/16:9, a handmade clay kitchen with real fingerprint texture and one sound per beat. Tip: name the medium, give one sound per action, and check the beats before upscaling.
- •Vertical 9:16 product shot with on-screen callouts. Prompt: "Clean 3D product-render video, vertical 9:16, plain white studio background, single continuous shot, locked-off camera. A matte-black quadcopter drone rotates slowly on a turntable at the center of the frame. Floating callout labels appear one at a time with thin leader lines pointing to the drone: first 'CARBON FRAME', then '4K GIMBAL', then '38-MIN FLIGHT'. The drone and background stay steady; only the drone's rotation and the labels animate. No people, no photorealism. Sound: a soft tech whoosh as each label appears, a low steady hum underneath. No dialogue. Every word spelled exactly as given, no other text." At 5s/720p/9:16 the drone turns while three legible callouts appear in turn, each with a whoosh. Tip: set aspect_ratio to 9:16 and keep callouts at 720p or higher for sharp labels.
More prompts this model is built for: "isometric low-poly city block, cars loop around one roundabout"; "paper-collage explainer, three labelled icons fade in with a page-turn sound"; "[0-3s] a doodle rocket draws itself / [3-6s] it lifts off with a whoosh".
