GeminiModel and multi modal prompts

[dpolistwm]

Anyone has been successful? No luck here.

[Unshure]

Hi @dpolistwm, we have some docs on how to get started with the Gemini model provider: https://strandsagents.com/latest/documentation/docs/user-guide/concepts/model-providers/gemini/

Can you let me know what issues you are running into?

[yonib05]

Hi @dpolistwm, sorry this sat unanswered for so long after Unshure's follow-up, that's on us.

Since we never got details on what failed, here's the state of things: multimodal input does work with GeminiModel today. Image and document content blocks are converted to Gemini inline data parts, and we run integration tests against exactly this shape:

from strands import Agent
from strands.models.gemini import GeminiModel

model = GeminiModel(client_args={"api_key": "<KEY>"}, model_id="gemini-2.5-flash")
agent = Agent(model=model)

with open("photo.png", "rb") as f:
    image_bytes = f.read()

result = agent([
    {"text": "What do you see in this image?"},
    {"image": {"format": "png", "source": {"bytes": image_bytes}}},
])

Two things to check if it's still not working for you: bytes must be the raw file bytes (not a base64 string, not a file path or URL), and format must be one of png, jpeg, gif, webp for images. Documents work the same way with {"document": {"format": "pdf", "source": {"bytes": ...}}}.

The provider docs have since moved and now include full multimodal examples: https://strandsagents.com/docs/user-guide/concepts/model-providers/google/

If you hit a concrete error with a current SDK version, please open a bug with the traceback so we can dig in: https://github.com/strands-agents/harness-sdk/issues/new?template=bug_report.yml. Hope this gets you unblocked!