Tools · News
DeepMind's Gemini Omni prompt guide: direct the shot, then edit by talk
Google DeepMind's official Gemini Omni prompt guide (companion to Google's 19 May 2026 Omni launch) walks makers through framing, style, lighting, location and action, then conversational edits: camera angle and move, action changes, style shifts, storyboards, image/video/audio references, scene extensions in 10-second steps up to 40 seconds, and first-to-last frame paths. Compared with Veo, DeepMind says you need not spell every detail; Omni's world knowledge fills in. Camera language on the page includes oner, locked off, push in, punch in and dolly zoom. Checked 9 Nov 2026 against DeepMind's live guide; the scene-extension note still says Omni 1.1. Not a first-hand test.
What changed
Instead of rewriting a full Veo-style prompt for every tweak, DeepMind's guide shows a director's loop: set frame, style, light, place and action once, then change camera, action, style or references in plain language while Omni keeps what already works. Scene extensions run in 10-second steps up to 40 seconds; first and last frames can lock a continuous move.
Who it's for
Editors and AI film directors shaping Gemini Omni shots who want camera grammar, conversational revise passes, storyboards and multi-input references from the maker's own page rather than a third-party roundup.
What it means for a cut
Treat it as vendor craft notes, not a tested AFN recipe. The live guide's scene-extension line still names Omni 1.1 as of the 9 Nov 2026 check, so confirm behaviour on the Omni build you actually use. Access, length caps and credit costs sit on Google's product pages, not in this guide.
Tags: google, deepmind, gemini-omni, prompts, camera, video, storyboard
Related notes
- Google Flow adds five artist-built tools, including Pocket Studio for multi-angle phone takes Google Labs, 12 Nov 2026
- Media Composer gets Ready to Play: Veo and Gemini clips straight into your bin Avid, 10 Nov 2026
- Formula E Creators Challenge GEN4 / Formula E, 15 Sep 2026
Our note credits Google DeepMind and links to the original; we do not republish the piece. More from May 2026, or the latest on News.