The $10,000 problem: Why enterprise video rarely gets made
For most enterprises, a 90-second training video or a product explainer has never been an easy ask. According to a 2025 survey by the Corporate Executive Board, the average internal video production costs between $8,000 and $15,000 per finished minute, with timelines stretching from two to six weeks. The process typically involves drafting a detailed brief, hiring an internal film crew or an external vendor, scheduling a shoot, editing the footage, and then going through multiple rounds of revisions with stakeholders. A single legal review that changes a line of on-screen text can force the entire chain to restart, adding another week and thousands of dollars. As a result, according to Gartner, nearly 60% of planned internal videos are never completed, leaving employees without critical training or product knowledge.
Gemini Omni Flash: A conversational approach to video creation
Google is aiming to rewrite that equation with Gemini Omni Flash, the first model in its new 'Omni' family, now available via the Gemini API. Unlike traditional video production tools that rely on clunky timelines and manual editing, Omni Flash treats the entire process as a real-time conversation. A user simply describes the video they need — for example, 'Create a 60-second training video on data privacy compliance' — and the model generates a rough cut, complete with voiceover, visuals, and transitions. The user can then refine the output through natural language commands, such as 'Change the background to a corporate office' or 'Add a slide about GDPR penalties.' Google claims the model can produce a first draft in under three minutes, compared to the industry average of several days.
How the technology works: Multimodal understanding and generation
Gemini Omni Flash builds on Google's previous generation of Gemini models but introduces a unified architecture that processes audio, video, and text inputs simultaneously. According to Google's technical documentation, the model uses a novel 'cross-attention' mechanism that aligns visual frames with spoken narration in real time, allowing it to synchronize lip movements, background visuals, and on-screen text. For example, if a user requests a video with a specific statistic appearing on screen at a certain timestamp, Omni Flash can generate that exact frame without needing a separate rendering step. The model also supports dynamic content updates: if a legal review changes a compliance regulation, the user can say 'Update the third slide with the new 2026 data privacy rules,' and the model will regenerate only the affected segment, rather than the entire video.
Competitive landscape: Google vs. OpenAI, Meta, and startups
Google is not alone in targeting enterprise video production. OpenAI's Sora, launched in early 2025, offers text-to-video generation but has been criticized for lacking fine-grained control over specific scenes and audio synchronization. Meta's Make-A-Video, meanwhile, focuses on consumer-facing short clips rather than corporate training. Startups like RunwayML and Synthesia have also carved out niches: Synthesia reported $120 million in annual recurring revenue in 2025, driven by its avatar-based video creation platform. However, none of these competitors offer the conversational, iterative refinement that Omni Flash promises. According to a Google spokesperson, the key differentiator is that Omni Flash treats the video as a 'living document' that can be edited through natural dialogue, rather than a one-shot generation. Analysts at IDC estimate that the enterprise video AI market could reach $4.5 billion by 2027, making this a high-stakes battleground.
Implications for the enterprise: Cost savings, but also new risks
The potential benefits for enterprises are substantial. A company that previously spent $50,000 on a series of training videos could now produce the same content for a fraction of the cost, perhaps under $1,000 in API usage fees. However, the technology also introduces new risks. AI-generated videos may contain factual errors, biases, or non-compliant content that could expose companies to legal liability. For instance, a training video on workplace harassment that inadvertently uses biased language could lead to regulatory fines. Google has implemented a 'safety filter' that flags potentially problematic content, but the company acknowledges it is not foolproof. Additionally, the ease of production could lead to a flood of low-quality internal videos, overwhelming employees. As one industry analyst noted, 'The barrier to video production is dropping from a high wall to a speed bump, which is great for productivity, but it also means companies need new governance policies.'