The GenaiVertexV1beta1TranscriptionConfig now includes a mode field; the SMART mode performs disfluency removal (filler words, repetitions) and grammatical cleanup, though it is incompatible with timestamps and diarization. For speech generation, the array-based speechConfig has been replaced by structuredSpeechConfig, utilizing the new GenaiVertexV1beta1SpeakerConfig to support multi-speaker setups.
Video generation capabilities expand with the EXTEND task in GenaiVertexV1beta1VideoConfig. Furthermore, GenaiVertexV1beta1VideoContent introduces processingType, enabling AGENTIC dynamic navigation (the new default for Gemini 1.5 Pro+ models) as an alternative to STATIC fixed-rate frame extraction. Developers can also now specify video output resolutions up to 4K. Finally, the base_agent field in the Agent resource is now immutable, requiring a new agent creation for base model changes.
