Developers can now fine-tune video generation experiments using new LoRA (Low-Rank Adaptation) configurations via CloudAiLargeModelsVisionGenerateVideoExperiments.loraConfig. This includes specifying multiple adapters with id and weight. Additionally, superResConfig allows for Pro Super Resolution upsampling with modes like MODE_4K and MODE_8K. For Vertex AI Agents, the new GoogleCloudAiplatformV1Agent.modelConfig property enables setting a default model and temperature for the agent's tasks. Finally, GoogleCloudAiplatformV1EvaluationRunEvaluationConfig now includes allowCrossRegionModel to permit evaluation runs to utilize models across different regions.
Medium Impact
DocsAdvanced Video Gen & Agent Model Control
Modified API Methods (2):
