Google’s new anything-to-anything AI model is wild

2026-05-23 · Source: The Verge · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Novice, medium

Summary

Google has released Omni Flash, the first model in its new Omni family of generative AI models, now available in its AI video generation and editing platform, Flow. Omni Flash improves upon the previous Veo model by allowing video uploads as starting points alongside text prompts, incorporating more real-world knowledge, and better character consistency. While some results show significant improvement, particularly in consistency, "AI jump scares" and object inconsistencies persist. The service operates on a credit system, costing 15-40 credits per video and 40 credits per edit, with a a \$20-per-month AI Pro plan offering 1,000 credits. Omni Flash also demonstrates a purported strength in adding AI-generated elements to real videos, producing highly convincing deepfakes that can fool observers, despite minor glitches and an uncanny valley effect.

Key takeaway

For creative technologists exploring generative AI for video content, Google's Omni Flash offers advanced capabilities for creating and editing videos, including highly realistic deepfakes. Be aware that achieving a precise vision can be costly due to the credit system and iterative editing. You should carefully weigh the balance between creative control, output quality, and the potential for uncanny valley effects, especially when integrating AI elements into real footage.

Key insights

Google's Omni Flash significantly advances AI video generation, creating highly convincing deepfakes despite persistent inconsistencies and cost implications.

Principles

• AI video generation tools are surprisingly good and easy to use.
• Character consistency remains a challenge for current models.
• AI-generated elements in real videos can be highly deceptive.

Method

Users upload a video and/or text prompt to Google's Flow platform, then generate or edit videos. Edits are also text-prompt driven.

In practice

• Generate videos from existing footage and text prompts.
• Create convincing deepfakes from selfie videos.
• Edit generated videos using natural language commands.

Topics

• Generative AI
• Video Generation
• Google Omni Flash
• Deepfakes
• AI Video Editing
• Uncanny Valley

Best for: AI Engineer, Machine Learning Engineer, Computer Vision Engineer, Creative Technologist, Tech Journalist, General Interest

Related on AIssential

• Google's brand new video gen model: Omni — Omni is Google's new AI video model, generating long, consistent, photorealistic videos with conversational editing.
• Google launches Nano Banana 2 Lite for fast AI images and Gemini Omni Flash for video via API — Google's new AI models enable rapid, cost-effective image generation and API-driven video creation.
• Introducing Gemini Omni — Gemini Omni Flash enables multimodal video creation and conversational editing, grounded in real-world knowledge and physics.
• Google’s Gemini Omni Can Generate Videos With Shockingly Accurate Text 😳 — Gemini Omni demonstrates advanced video generation with accurate text, nearing indistinguishable realism despite minor flaws.
• DreamID-Omni: Unified Framework for Controllable Human-Centric Audio-Video Generation — DreamID-Omni unifies human-centric audio-video generation with disentangled control over identity and timbre.
• Google Omni made this hype video in less than 15 minutes — AI tools like Google Omni enable rapid, high-quality content creation for marketing and media.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Verge.