Gemini 3 Multimodal Prompts Across Video, Audio and Text

In Gemini 3, text, images, audio, and video should be treated as equal-class inputs. Reference each specific modality clearly in your prompt so the model synthesizes across them rather than analyzing them in isolation.

<context> <video>{{USER_SESSION_VIDEO}}</video> <audio>{{USER_THINK_ALOUD_AUDIO}}</audio> <logs>{{CLIENT_ERROR_LOG_TEXT}}</logs> </context> <task>Based on the video, audio, and text logs above, synthesize a unified timeline correlating spoken user frustration in the audio with visual UI freezes in the video and exception stack traces in the logs.</task>