Builds multimodal ChatMessageContentBlock[] and the matching
ReferenceMediaKind label for a single reference asset:
Image / Screenshot -> requires an image/* MIME type and returns a single image_url data URL block (referenceKind: 'Image').
Video -> requires a video/* MIME type. When videoCapable === true, the MIME type is video/mp4 or video/webm,
and bytes.byteLength <= MAX_INLINE_REFERENCE_VIDEO_BYTES, returns a single video_url data URL block (referenceKind: 'Video (native)').
Any other video/* -> calls extractKeyframes (MaxFrames: 16, BuildContactSheet: true)
and returns up to 4 contact-sheet image_url blocks (or the first 4 individual frame
image_url blocks when no contact sheets are returned) with referenceKind: 'Video (keyframes)'.
Builds multimodal
ChatMessageContentBlock[]and the matching ReferenceMediaKind label for a single reference asset:Image/Screenshot-> requires animage/*MIME type and returns a singleimage_urldata URL block (referenceKind: 'Image').Video-> requires avideo/*MIME type. WhenvideoCapable === true, the MIME type isvideo/mp4orvideo/webm, andbytes.byteLength <= MAX_INLINE_REFERENCE_VIDEO_BYTES, returns a singlevideo_urldata URL block (referenceKind: 'Video (native)').video/*-> callsextractKeyframes(MaxFrames: 16,BuildContactSheet: true) and returns up to 4 contact-sheetimage_urlblocks (or the first 4 individual frameimage_urlblocks when no contact sheets are returned) withreferenceKind: 'Video (keyframes)'.