natolambert/gemini-feedback

Gemini Feedback

Get feedback from Gemini API on a diagram image.

Usage

/gemini-feedback <path-to-image> [context]

Instructions

When this command is invoked:

1. Read the image file at the provided path
2. Call the Gemini API to get feedback on the diagram
3. Use this Python snippet to call Gemini (uses the newer google-genai package):

import base64
import os
from google import genai

client = genai.Client(api_key=os.environ['GEMINI_API_KEY'])

with open('<IMAGE_PATH>', 'rb') as f:
    img_bytes = f.read()

prompt = '''Review this diagram for a textbook. Be concise and specific.

Context: <CONTEXT>

Please provide:
1. Overall assessment (1-2 sentences)
2. Specific issues to fix (be detailed about visual problems like alignment, overlapping, etc.)
3. Suggestions for improvement
'''

response = client.models.generate_content(
    model='gemini-2.5-flash',
    contents=[
        prompt,
        genai.types.Part.from_bytes(data=img_bytes, mime_type='image/png'),
    ],
)
print(response.text)

4. Replace <IMAGE_PATH> with the actual path provided
5. Replace <CONTEXT> with any context provided, or use "Technical diagram for ML/AI textbook"
6. Ensure GEMINI_API_KEY is set in environment
7. Run using: uv run python -c "..."
8. Present the feedback to the user

Example

/gemini-feedback diagrams/generated/png/tool_use_generation.png "Diagram showing tool use interleaving in LLM generation"