Veo 3.1 Ingredients to Video: More consistency, creativity and control

Google DeepMind pushed out a fairly significant update to Veo 3.1's "Ingredients to Video" feature this week, and the headline change is consistency. Feed the model a handful of reference images — a character, a texture, a background — and it now keeps those elements looking the same across multiple scenes, instead of the identity drift that's plagued a lot of AI video tools since day one.

The more practical addition, though, is native 9:16 vertical output. Up until now, anyone wanting a Shorts-ready or TikTok-ready clip had to generate horizontal video and crop it, losing detail in the process. Now Ingredients to Video renders straight to portrait, and Google is shipping it directly inside YouTube Shorts and the YouTube Create app, not just buried in a developer console somewhere.

Resolution got a bump too. Veo 3.1 now upscales to 1080p and 4K, aimed squarely at people doing actual production work rather than just messing around for fun. That tier lands in Flow, the Gemini API, and Vertex AI — the enterprise-facing side of the house — while casual users in the Gemini app get the expressiveness and portrait-mode improvements without the resolution ceiling raised as high.

Google is also leaning on Nano Banana Pro, its Gemini 3 Pro Image model, as the suggested way to generate those ingredient images before feeding them into Veo. It's a small detail, but it signals where Google wants this workflow to live: entirely inside its own stack, from stills to motion to final export.

On the provenance side, every clip still carries the SynthID watermark, and Google's video-verification tool in the Gemini app — expanded back in December — lets anyone upload a clip and ask whether it came from Google's models. Useful, sure, but it only catches Google-made content, which is a fairly narrow slice of what's about to flood short-form platforms.