Google's annual developer event is next week, and something unusual has already leaked out. A video tool called Gemini Omni showed up in the Gemini app for a small number of users over the weekend, with the interface inviting them to remix videos, edit clips by chatting, and try pre-made templates. This was either a testing slip or a controlled preview. Either way, it tells us something real about where Google is heading with AI video.
The first thing to understand is who Google is competing against. ByteDance, the company behind TikTok, released a video model called Seedance 2.0 in February 2026. It generates up to 15 seconds of video with audio already synchronized, handles complex motion scenes, and produces what many observers describe as near-cinematic quality. Disney sent ByteDance a cease-and-desist over alleged training data violations, and US senators have called for it to be shut down. But the underlying product is widely regarded as the current benchmark for AI video quality.
By that visual quality measure, Gemini Omni is not there yet. Early test videos showed solid results for simple scenes but obvious glitches in more complex ones, with objects appearing and disappearing in ways that break realism. Google appears to know this, and seems unbothered by it.
The reason is that Google has run this exact playbook before. When it launched its AI image tool inside Gemini, it debuted with middling visual quality scores but topped charts for editing capability. It was later upgraded into a genuinely competitive image system. The pattern here looks identical: launch with strong editing features, then improve generation quality as compute costs fall and the model matures.
What makes this commercially interesting is that editing capability is actually what most businesses need. Removing a watermark from a clip, swapping a product in a scene, adjusting a line of dialogue without reshooting, rewriting a scene to match updated messaging: these are real, repeatable problems for marketing teams, training departments, and communications staff. Generating a cinematic video from scratch is impressive. Being able to fix and adapt existing video in a chat window is what gets used every day.
The business context around AI video is already serious. Traditional video production runs around $4,500 per minute. AI-assisted production brings that closer to $400 per minute, and the time to produce a 60-second marketing video has reportedly dropped from 13 days to roughly 27 minutes. Enterprise spending on AI video tools grew 127% year-over-year in 2025. Around 73% of Fortune 500 companies have already integrated AI video tools into content workflows.
For Google, the timing is also strategic. OpenAI effectively stepped back from its Sora video tool earlier this year, leaving a clear opening. ByteDance's Seedance faces legal pressure from major studios and US regulatory scrutiny, which makes it less viable for risk-conscious enterprises. A Google-branded video tool, built into a product many businesses already pay for, arrives at a moment when the competition is either retreating or under legal fire.
The credit limits spotted in early testing are worth noting too. Two short video prompts burned through nearly all of one user's daily allocation on a paid plan. That is a pricing signal: video generation is expensive to run, and Google intends to charge accordingly through tiered plans. Hints point to at least two versions, a faster lighter-weight option and a higher-quality one, similar to how the rest of Google's AI lineup is structured.
The full announcement comes next week. What is already clear is that Google's AI video strategy is not built around winning a quality benchmark at launch. It is built around being embedded where people already work, offering editing capabilities that serve real daily tasks, and improving the underlying quality from there.