Ag2 Multimodal Input

Multimodal inputs

When to use

The user wants the agent to process non-text input: an image to describe, audio to transcribe, video to summarise, or a PDF / document to extract from. The same factory pattern works across providers; per-provider support varies.

60-second recipe

from autogen.beta import Agent
from autogen.beta.config import GeminiConfig
from autogen.beta.events import ImageInput

agent = Agent(
    "vision",
    "You describe images.",
    config=GeminiConfig(model="gemini-3-flash-preview"),
)

image = ImageInput("https://example.com/photo.jpg")
reply = await agent.ask("Describe this image in detail.", image)
print(reply.body)

Multiple inputs in one ask are fine:

reply = await agent.ask(
    "Compare these two images.",
    ImageInput("https://example.com/before.jpg"),
    ImageInput("https://example.com/after.jpg"),
)

Input factories

Each accepts the same four data sources:

from autogen.beta.events import ImageInput

ImageInput("https://example.com/photo.jpg")     # URL
ImageInput(path="photo.jpg")                    # local file
ImageInput(data=raw_bytes, media_type="image/png")  # bytes
ImageInput(file_id="file-abc123")               # provider-uploaded

Provider matrix

Unsupported combinations raise UnsupportedInputError with a clear message.

Gemini has the broadest multimodal support. If you don't know which provider to pick for a multimodal task, start there.

Provider-specific niceties

Gemini — YouTube URLs work directly

from autogen.beta.events import VideoInput

video = VideoInput("https://www.youtube.com/watch?v=dQw4w9WgXcQ")
reply = await agent.ask("Summarize this video.", video)

Gemini — large files (> 20MB) via Google Files API

from google import genai
from autogen.beta.events import VideoInput
import time

client = genai.Client()
uploaded = client.files.upload(file="large_video.mp4")
while uploaded.state.name == "PROCESSING":
    time.sleep(2)
    uploaded = client.files.get(name=uploaded.name)

video = VideoInput(uploaded.uri)

Gemini — vendor_metadata

ImageInput(data=raw, media_type="image/jpeg", vendor_metadata={"media_resolution": "MEDIA_RESOLUTION_LOW"})

VideoInput(path="lecture.mp4", vendor_metadata={
    "video_metadata": {"start_offset": "60s", "end_offset": "120s", "fps": 0.5},
})

OpenAI — image detail

ImageInput(data=raw, media_type="image/png", vendor_metadata={"detail": "low"})  # "low" | "high" | "auto"

Anthropic — File ID + prompt caching

import anthropic
from autogen.beta.events import ImageInput, DocumentInput

client = anthropic.Anthropic()
uploaded = client.beta.files.upload(file=("photo.jpg", open("photo.jpg", "rb"), "image/jpeg"))

# filename determines block type (image vs document)
image = ImageInput(file_id=uploaded.id, filename="photo.jpg")

# Cache an attachment so subsequent turns skip re-uploading
doc = DocumentInput(path="report.pdf", vendor_metadata={"cache_control": {"type": "ephemeral"}})

FilesAPI — upload lifecycle, provider-agnostic

For any provider that has a file API (OpenAIConfig, OpenAIResponsesConfig, AnthropicConfig, GeminiConfig):

from autogen.beta import FilesAPI
from autogen.beta.config import OpenAIResponsesConfig

files = FilesAPI(OpenAIResponsesConfig(model="gpt-5-mini"))

uploaded = await files.upload(path="report.pdf", purpose="assistants")
print(uploaded.file_id)

# Or from bytes (filename required)
uploaded = await files.upload(data=b"...", filename="hello.txt", purpose="assistants")

# List, read, delete
all_files = await files.list()
data = await files.read(uploaded.file_id)        # NotImplementedError on Gemini
await files.delete(uploaded.file_id)

Pass the file_id to DocumentInput, ImageInput, etc.:

from autogen.beta.events import DocumentInput

doc = DocumentInput(file_id=uploaded.file_id)
reply = await agent.ask("Summarize this report.", doc)

Going deeper

• website/docs/beta/inputs/inputs.mdx — full provider matrix and vendor_metadata reference.
• website/docs/beta/advanced/files.mdx — FilesAPI reference (upload / list / read / delete).
• For tools that return images / binary back to the LLM, see ag2-add-custom-tool (ImageInput, BinaryInput, ToolResult).

Common pitfalls

• Picking a provider that doesn't support your input type — silently you'll get UnsupportedInputError. Check the matrix; Gemini is broadest.
• `FilesAPI.read()` on Gemini — raises NotImplementedError. Gemini doesn't expose download.
• Calling `files.upload(data=...)` without `filename=` — raises ValueError. Filename is required for in-memory uploads.
• Supplying more than one source to a factory — not an error. The factory resolves in priority order url > file_id > path > data, so extra sources are silently ignored. Pass exactly one to get what you intend. Supplying zero sources raises ValueError.
• Anthropic `ImageInput(file_id=...)` without `filename=` — Anthropic decides block type (image vs document) by filename extension. Pass it.
• Gemini `vendor_metadata` keys are nested — video_metadata itself takes a dict. Check the doc table for shape.
• Forgetting to wait for Gemini file processing — large uploads have a PROCESSING state. Poll client.files.get(name=...) until ready before referencing the URI.