Google Drops Video Gen Prices to 7 Cents a Second Because Hollywood Was Too Expensive

Google DeepMind rolled out Gemini Omni 1.2 Flash, officially moving the Omni video generation family out of experimental preview and giving developers actual commercial SLA guarantees. The headline feature isn't even a mind-blowing visual leap, but a price cut down to just 0.07 dollars per second of 720p footage, which undercuts the previous version by thirty percent. Alongside the discount, the maximum clip length expanded from ten to fifteen seconds with one-second increments. This brings the cost of a full minute of raw video down to about 4.20 dollars, making automated ad creation considerably cheaper for marketing teams churning through dozens of alternative creative options.

API billing relies on token consumption, charging 12 dollars per million output tokens and 1.50 dollars for input data like text, images, or short video snippets. Free tiers are predictably nonexistent, and legacy version 1.1 requests automatically route to the fresh pricing structure without requiring developers to rewrite their codebases. Under the hood, the model accepts prompts, character images, product references, style guides, and audio cues, promising better object consistency and improved lip-syncing for multiple speakers. Editing capabilities also allow users to modify backgrounds, lighting, weather, or aspect ratios in a conversational loop without regenerating everything from scratch. Complex physical simulations and fast sports action still occasionally break the engine, and the API officially speaks only English for now.

Early performance claims rely entirely on internal DeepMind evaluations boasting a sixty-two percent win rate in visual consistency over the predecessor, though independent benchmarks are notably absent. Meanwhile, the enterprise partner lineup points straight toward the infrastructure market, with platforms like Runway, Krea, Picsart, and QuickFrame jumping on board to pipe the tech directly to millions of creators. Picsart alone plans to expose the model to its massive user base, while QuickFrame leans on the new fifteen-second limit to generate standard TV commercial lengths in a single 1080p pass.

At seven cents a second, building an in-house GPU cluster for video generation suddenly makes financial sense only for massive enterprise volumes or strict data paranoia. Google clearly decided that fighting for prestige in cinematic quality is far less profitable than becoming the invisible plumbing for every ad agency on the planet.

Source: Google Blog