Google's Veo 3.1 is better at generating videos from images (www.engadget.com)

Google has released Veo 3.1, an incremental but practical upgrade to its video-generation model that’s available now through the Gemini API and integrated into Google’s Flow video editor. Building on Veo 3 (announced at Google I/O 2025), Veo 3.1 improves prompt adherence and is better at using uploaded image “ingredients” to condition generated video. New capabilities include converting still images into motion and generating audio alongside video — something Veo 3 couldn’t do — plus a Flow “Frame to Video” feature that interpolates from a user-supplied first and last frame to produce the in-between footage (and audio). The model also powers editable tasks like extending clips and inserting objects into existing footage. For the AI/ML community, Veo 3.1 signals continued focus on multimodal controllability and practical editing workflows rather than raw novelty. Technically, it tightens alignment between text and image conditioning, supports simultaneous video+audio synthesis, and offers finer temporal control through frame interpolation — all useful for content-creation pipelines and editor integrations. Samples still show variable, sometimes uncanny realism compared with rivals like OpenAI’s Sora 2, but Google’s emphasis on usability for editors (and parity with features in Adobe Firefly) highlights progress toward production-ready generative video tools and raises new evaluation and safety considerations for multimodal models.