Google DeepMind

Native vertical (portrait) video output for Ingredients-to-Video. Stronger reference-image guidance (up to multiple images) to lock characters, textures, and backgrounds. Higher-quality upscaling options that reach 1080p and 4K for more production-ready results. More expressive motion and better controllability even from simpler prompts. Rollout tied to creator workflows, including integration into YouTube Shorts and the YouTube Create app.

Richer native audio generation: supports natural conversations, synchronized sound effects, ambient audio, and better mixing. Better understanding of cinematic styles, giving users more control over the narrative tone and visual storytelling. Stronger prompt adherence: generated videos more faithfully reflect input prompts, especially from images. Character consistency across scenes: visual consistency in appearance (e.g., faces, objects) is improved over multi‑shot sequences. Reference image conditioning (“Ingredients to video”): up to 3 reference images can be provided (character, object, scene) to guide style or consistency. Scene extension: existing video clips can be extended by generating new segments that tie into the last second of preceding video, preserving visual continuity. First → Last frame transitions: you can provide a starting frame and an ending frame, and Veo 3.1 will generate a smooth bridging animation (with audio) between them.

Native Audio Generation: Generates synchronized audio (dialogue, sound effects, music) with video. High Realism: Produces lifelike visuals, realistic physics, accurate human features, and lip-sync. Cinematic Camera Controls: Offers manual control over camera angles, motion, and perspectives. Scene Builder: Allows editing/extending shots, maintaining motion and character consistency. Asset Management: Organize, reference, and reuse characters, props, and prompts across scenes. Prompt Adherence: Follows complex, detailed text/image prompts accurately. Seamless Transitions: Maintains continuous motion and consistent characters between shots. Learning Tools: Flow TV shows user-generated clips with prompts for inspiration and learning. Deepfake Mitigation: Embeds invisible watermarks to identify AI-generated videos.

Enhanced Realism and Fidelity: Significantly better detail, realism, and reduction of visual artifacts. Advanced Motion Representation: Improved physics simulation for accurate, lifelike movements. Higher Output Quality: Supports 4K resolution, surpassing Veo 1’s lower-resolution capabilities. Expanded Camera Control: Offers a wider range of shot styles, angles, and movements. Broader Visual Styles: Faithfully follows complex and intricate prompts with greater precision. Improved Benchmark Performance: Outperforms Veo 1 in human evaluations for prompt adherence and overall preference. Cinematic Capabilities: Better suited for dynamic scenes, complex lighting, and artistic storytelling.