📰 Key Takeaways

Google launched Gemini Omni 1.2 Flash today, now available across the Gemini API and Gemini for Enterprise. The core of this update is native 15-second video generation, multi-turn conversational editing, and stronger adherence to reference materials. Pricing is 30% lower than Omni 1.1, at $0.07 per second for 720p video. Cost control extends into the workflow too: you can draft quickly at 360p or 720p, fix a single detail with follow-up edits instead of regenerating the whole clip, and then upscale the version you like to 1080p or 4K.

Video length goes up from 10 seconds in Omni 1.1 to 15 seconds, and you can set the duration anywhere from 3 to 15 seconds in 1-second steps. That’s enough to fit a product explainer or a complete social ad into a single generation. On the reference side, you can combine character, product, or style images, motion clips up to 3 seconds long, and audio prompts all at once. Faces, clothing, packaging typography, and brand colors stay consistent across different camera angles. In the official demo, a single model photo plus front and back shots of three outfits produced a 15-second virtual try-on video, with a 360-degree rotation and match cuts, while preserving the model’s identity and each garment’s cut, drape, and back-side details.

Consistency has also improved. Characters, objects, and scenes show less visual drift or warping from the first frame to the last, whether you’re interpolating between first and last keyframes or extending a clip. For editing, you can refine the same video turn by turn, such as swapping the product or background, changing the lighting or weather, or switching the aspect ratio, and multi-turn editing works across the full 15-second clip. See the original post for the rest of the details.

💬 JudyAI Lab Take

Google’s Gemini Omni 1.2 Flash moves video generation from “bet on the whole clip in one shot” to “fix it in stages.” Clips run up to 15 seconds, the price is 30% lower than the previous generation, and 720p costs $0.07 per second. AI builders should pay attention.

What has always hurt most with generative video is that one wrong detail means redoing the whole clip, which doubles both cost and time. The design thinking here is clear: draft at 360p or 720p, then use multi-turn conversation to fix just one detail, like swapping the product, changing the lighting, or switching the aspect ratio, and only upscale to 1080p or 4K at the very end. It’s basically bringing the software habit of “build a prototype first, then iterate” into video creation. Another thing worth noting is how reference materials can be combined. Character images, product shots, 3-second motion clips, and audio prompts can all go in together, and faces, clothing, and brand colors stay consistent across camera angles. We think this signals that the competition is shifting from “how good does it look” to “can you control it reliably, edit just one region, and save the cost of rerunning.”

If you’re planning a video workflow, try breaking down your own production steps first. Find which stages can be validated with low-resolution drafts, then decide where high-cost output is actually worth paying for.

📅 Source Info

🔗 Further Reading