Sending an Image to Gemini

Where we left off

Lesson 2 established the problem: some facts only exist in an image. This lesson is the smallest possible step toward fixing it, sending Gemini one image, alongside a text prompt, in a single generate_content call. No embedding, no retrieval yet, just: can Gemini look at a picture and answer a question about it?

The code, piece by piece

from google.genai import types

image_bytes = image_path.read_bytes()
image_part = types.Part.from_bytes(data=image_bytes, mime_type="image/png")

generate_content's contents argument isn't limited to a string, it accepts a list of Part objects, and Part isn't limited to text either (confirmed against this project's installed google-genai version, google/genai/types.py). Part.from_bytes wraps raw image bytes and a MIME type into the same kind of object a text prompt becomes internally; Gemini's models are natively multimodal, so mixing an image Part and a text Part in one contents list is not a special mode, it's the normal way to send more than plain text.

response = client.models.generate_content(
    model=CHAT_MODEL,
    contents=[image_part, QUESTION],
)

A bare string in contents (like QUESTION here) is automatically wrapped into a text Part for you, the SDK doesn't require you to write types.Part.from_text(text=...) yourself for something this simple. Order in the list matters for how the model reads the prompt, not for correctness, image-then-question and question-then-image both work; this course puts the image first throughout, so the question that follows always reads like "given this, answer: ...".