Gemini Omni 1.2 Flash: Google's New AI Video Model Is INSANE

Try Unlimited Seedance 2.5 on OpenArt: https://openart.ai/home?utm_source=youtube&utm_medium=influencer&utm_campaign=infl-youtube--na-acq-web&ref=BitBiasedAI-sd25 Google just launched Gemini Omni 1.2 Flash, an AI video model that can generate a 15-second video with sound for about $1.05. It's 30% cheaper than Omni 1.1, supports conversational video editing, and promises more accurate text, better camera control, and improved character consistency. But there's a catch. An independent tester called it the most instruction-exact AI video model he'd used, while also reporting that roughly one in four instructions gets skipped. So how good is Google's new AI video generator when you move beyond carefully selected demos? Google announced Gemini Omni 1.2 Flash on November 5, bringing its video generation technology into general availability through the Gemini API and Google Cloud. The model supports clips from 3 to 15 seconds, generation at 360p and 720p, and upscaling to 1080p or 4K. At just $0.07 per second for 720p generation, Google's pricing makes AI video experimentation significantly more affordable. But the cost of generating a clip isn't necessarily the cost of producing a finished video. Google says Omni 1.2 Flash can edit existing videos through natural-language instructions. Want to change a sunset into a nighttime scene? Replace an object without changing the background? Add rain while preserving the original camera movement? That's the promise behind conversational AI video editing. The model also supports image references, motion references, first-and-last-frame guidance, synchronized dialogue, lip-sync, and more controlled cinematic camera movements. But how much of that actually works? We examine Google's official demonstrations, including a continuous drone-style flight through the Pantheon, a dramatic shoreline zoom-out, an octopus-to-dancer transformation, animated science explainers with readable labels, and multi-character dialogue. Then we put Gemini Omni 1.2 Flash through increasingly difficult video generation challenges. We start with a cinematic forest scene featuring two deer, realistic lighting, smooth camera movement, and generated audio. Next comes a science explainer testing whether the model can correctly spell and position labels like Sunlight, Refraction, and Rainbow. From there, we explore conversational editing with a red sports car, transforming a desert sunset into a nighttime scene before adding heavy rain and reflections. We also test character consistency using reference images, checking whether the same adventurer maintains her facial features, clothing, and appearance across separate video generations. Finally, we push into complex physics and multi-step instructions: falling dominoes, moving objects, synchronized dancers, and a dramatic astronaut-to-alien-temple camera movement. These challenges target some of the exact weaknesses Google acknowledges, including fast action, complex physics, and rapid scene changes. Independent testing by Rohit Rao at Segmind adds another perspective. Across 20 launch-day tests costing $16.17, he praised the model's instruction accuracy while identifying problems with skipped instructions, floaty human motion, counting, and numerical chart accuracy. And the competition matters. Seedance 2.0 and Kling 3 remain important alternatives. According to Rao's assessment, Seedance 2.0 still performs better for photorealistic people, cinematic footage, and physics-heavy action, while Omni 1.2 Flash shows particular promise for typography, explainers, chart animations, and precise edits. We also examine the real economics of AI video generation. A single 15-second clip costs approximately $1.05 at 720p, but retries, corrections, editing passes, and upscaling can make the final production considerably more expensive. So is Gemini Omni 1.2 Flash a breakthrough for AI video creation, or a specialized tool that excels at following certain instructions while struggling with cinematic realism? And should creators, marketers, developers, and businesses choose Google's model over Seedance and Kling? Watch the full comparison to see where Gemini Omni 1.2 Flash delivers, where it struggles, and what Google's approach means for the future of AI video generation. 💬 What's the hardest prompt you'd want Gemini Omni 1.2 Flash to attempt? Drop it in the comments! CHAPTERS 00:00 Google's $1 AI Video Model: What's the Catch? 00:46 What Google Actually Changed Since Omni 1.1 03:55 Seedance 2.5 by OpenArt 05:58 The Demos Google Picked, and What's Missing From Them 07:30 Round One: A Simple Shot, Then a Text Challenge 09:09 Editing a Video by Talking to It 10:23 Can It Keep a Character Looking Like the Same Person? 11:19 Pushing Into Physics and Chaos 12:48 The Dollar Is the Price of One Try, Not One Finished Video 14:19 So Who Should Actually Use It? #GoogleAI #GeminiOmni #AIVideoGeneration #Seedance #ArtificialIntelligence