The video generation API receives significant experimental updates. A new outpaintConfig allows for video outpainting with specific resolution and frame count targets (e.g., 1920x1072 at 72 frames). The audioControl configuration enables developers to either reuse audio from an input video or provide a new target audio track via CloudAiLargeModelsVisionGenerateVideoRequestAudio. Additionally, a videoTransform schema has been introduced to handle noise strength and masking for video editing tasks, deprecating the older videoTransformMaskGcsUri and videoTransformStrength fields. Outside of video, the AudioTranscriptionConfig now deprecates adaptationPhrases in favor of custom_vocabulary, and the agents.list method has been clarified to only return resources owned by the specific calling end user.
Medium Impact
Docs