Shortly after Google I/O 2026, Google published nine demos that put its two flagship announcements to work: Gemini Omni for video generation and Gemini 3.5 Flash for agents and coding. The divide is clear — Omni creates, 3.5 Flash executes.
Gemini Omni: Video From Any Input
Gemini Omni turns any reference — an image, text, video, or audio — into a single cohesive output. Three of the demos focus on what the model calls reimagining the action: remaking a sculpture out of bubbles, dimming a room’s lights, or dropping a recursive glass-sphere world into an existing shot. Every video ships with an imperceptible SynthID digital watermark.
Editing Through Conversation
The standout Omni demos show multi-turn editing. A violinist can be transported to a new environment, have the violin made invisible, and shift the camera over the shoulder — each instruction building on the last while the character stays consistent and the physics hold up.
Gemini 3.5 Flash Takes on Agentic Tasks
- Auto-renames and categorizes unstructured asset libraries via Antigravity, sustaining frontier performance under supervision
- Generates several UX approaches for a checkout flow in about 60 seconds in AI Studio
- Is now the default model in the Gemini app and AI Mode in Search globally
One demo shows 3.5 Flash powering an information agent that watches for athlete sneaker drops, sending synthesized updates with links. Another has Search build an interactive explainer of Gyroid patterns, and a third builds a custom fitness tracker — both custom generative UI assembled on the fly.
Gemini Spark, Running on the Same Stack
The final demo introduces Gemini Spark, a personal agent that runs 24/7 on Gemini 3.5 and the Antigravity harness, integrated with Gmail, Docs, and Slides — demonstrated creating a list of nut-free snacks and adding them to an Instacart cart. Spark rolled out to Google AI Ultra subscribers in the US.
What It Means
Taken together, the nine demos outline Google’s 2026 strategy: Omni for generative creativity and 3.5 Flash for reliable action, both coordinated through Antigravity. For developers, the message is that video generation and agentic coding now share one platform — and one subscription tier.