Developers can now fine-tune video generation experiments using CloudAiLargeModelsVisionGenerateVideoExperiments. This includes the new loraConfig for applying LoRA adapters for Omni editing/stylization, and superResConfig for Pro Super Res upsampling with modes like MODE_4K and MODE_8K. For agents, the GoogleCloudAiplatformV1beta1Agent resource now supports an optional modelConfig field, allowing you to set default model and temperature parameters for an agent's generation tasks. Additionally, GoogleCloudAiplatformV1beta1AgentAnomalyDetectionScope introduces autoEnroll for easier agent monitoring. Evaluation workflows are enhanced with allowCrossRegionModel in EvaluateInstancesRequest and EvaluationRunEvaluationConfig, providing more flexibility. Granular control over automated evaluation metrics is also available via stepAutoraterConfigs in PredefinedMetricSpec. Note that temperature and topP fields in GenaiVertexV1beta1GenerationConfig have been deprecated; developers should update their usage accordingly.
High Impact
Docs