Summary
Video results currently piggyback on FileOutput in MultimodalResponse.files[]. Add a dedicated VideoOutput type with video-specific metadata (duration, resolution, aspect_ratio, has_audio) and a videos property on MultimodalResponse.
Problem
Today, a generated video is stored as:
FileOutput(url="...", data=None, mime_type="video/mp4", filename="generated_video.mp4")
This loses video-specific metadata and forces users to check files[] for video content:
# Current — awkward
result = await app.ai_generate_video(...)
video_file = result.files[0] # Is this a video? A PDF? Who knows
Solution
# multimodal_response.py
@dataclass
class VideoOutput:
url: Optional[str] = None
data: Optional[bytes] = None
mime_type: str = "video/mp4"
filename: Optional[str] = None
duration: Optional[float] = None # seconds
resolution: Optional[str] = None # "1080p"
aspect_ratio: Optional[str] = None # "16:9"
has_audio: Optional[bool] = None
cost_usd: Optional[float] = None
async def save(self, path: str): ...
async def get_bytes(self) -> bytes: ...
# MultimodalResponse additions
class MultimodalResponse:
videos: List[VideoOutput] # NEW
has_video: bool # NEW propertyDeveloper Experience
result = await app.ai_generate_video(
"A cat playing",
model="openrouter/google/veo-3.1",
resolution="1080p",
)
# Clean access
print(result.has_video) # True
print(result.videos[0].duration) # 8.0
print(result.videos[0].resolution) # "1080p"
await result.videos[0].save("cat.mp4")
# Still backward compatible via files[]
print(result.has_file) # Also True (videos appear in files too)Dependencies
Files
| File |
Change |
sdk/python/agentfield/multimodal_response.py |
Add VideoOutput class, add videos/has_video to MultimodalResponse |
Acceptance Criteria
Notes for Contributors
Severity: LOW — Quality-of-life improvement.
Follow the existing ImageOutput / AudioOutput pattern exactly. The save() method should handle both URL download and raw bytes, same as the image/audio equivalents.
Summary
Video results currently piggyback on
FileOutputinMultimodalResponse.files[]. Add a dedicatedVideoOutputtype with video-specific metadata (duration, resolution, aspect_ratio, has_audio) and avideosproperty onMultimodalResponse.Problem
Today, a generated video is stored as:
This loses video-specific metadata and forces users to check
files[]for video content:Solution
Developer Experience
Dependencies
Files
sdk/python/agentfield/multimodal_response.pyVideoOutputclass, addvideos/has_videotoMultimodalResponseAcceptance Criteria
VideoOutputdataclass with url, data, mime_type, duration, resolution, aspect_ratio, has_audio, cost_usdVideoOutput.save()andVideoOutput.get_bytes()methods (mirror existingImageOutput/AudioOutputpattern)MultimodalResponse.videospropertyMultimodalResponse.has_videopropertyfiles[]for backward compatibilitypytest sdk/python/passesruff check sdk/python/passesNotes for Contributors
Severity: LOW — Quality-of-life improvement.
Follow the existing
ImageOutput/AudioOutputpattern exactly. Thesave()method should handle both URL download and raw bytes, same as the image/audio equivalents.