# Google's New Gemini Omni Is Too Much.

- Channel: [AI For Humans](https://www.youtube.com/channel/UCghJTNTO9kcDeUFXMuSDGLQ)
- Published: 2026-05-20
- Source video: [Watch on YouTube](https://youtube.com/watch?v=zc1_YAi7scA)
- Read on ReadTube: https://www.read.tube/p/videos/zc1_YAi7scA
- Language: Original (source language)

## Summary

**Google Unveils Gemini Omni**

Google unveiled Gemini Omni, a new video world model focused on better generation, editing, and multimodal reasoning. The hosts found its strongest use cases in highly consistent video-to-video edits, explainers, and style transfers, while warning that physics and cameo-style voice generation still have noticeable misses.

Google’s IO 2026 announcements centered on a broad push across AI video, search, and agentic tools, with Gemini Omni as the standout demo. The model is presented as a world model rather than a pure video generator, with early strengths in editing existing clips, preserving character consistency, and transforming scenes into new styles or settings. The hosts were most impressed by how it could keep a voice and visual identity stable across multiple edits, while also automatically changing backgrounds and props in surprisingly coherent ways.

They were less convinced by some physics and cameo-style features. Omni’s real-time avatar and voice mimicry seemed promising but uneven, and some tests broke down after repeated generations. Google also added content credentials verification to help identify AI-edited media, reflecting growing concern about deepfakes and synthetic content.

The episode also covered several adjacent launches:
- Gemini 3.5 Flash is a fast, capable lightweight model that benchmarks strongly for its class, though a true Pro model is still coming later.
- Gemini Spark is a cloud-based agent aimed at long-running, everyday tasks across email, calendar, and Drive.
- Ask YouTube promises AI-guided video search and chapter jumping.
- Docs Live turns spoken requests into live documents and can pull in related files to build drafts on the fly.
- Google Maps is gaining Genie 3-style interactive outputs, pointing toward AI-assisted navigation and exploration.

The hosts framed the bigger picture as a shift toward AI agents that can do real work, not just generate text or code, and they noted that Google seems increasingly serious about bundling models with practical tools. The episode ended with a major insider note: Andrej Karpathy is joining Anthropic, a move they see as a significant signal for frontier-model competition and recursive self-learning research.