Generate Images with Intermediate Steps

Code

from typing import Iterator

from agno.agent import Agent, RunOutput
from agno.models.openai import OpenAIChat
from agno.tools.dalle import DalleTools
from agno.utils.common import dataclass_to_dict
from rich.pretty import pprint

image_agent = Agent(
    model=OpenAIChat(id="gpt-5-mini"),
    tools=[DalleTools()],
    description="You are an AI agent that can create images using DALL-E.",
    instructions=[
        "When the user asks you to create an image, use the DALL-E tool to create an image.",
        "The DALL-E tool will return an image URL.",
        "Return the image URL in your response in the following format: `![image description](image URL)`",
    ],
    markdown=True,
)

run_stream: Iterator[RunOutputEvent] = image_agent.run(
    "Create an image of a yellow siamese cat",
    stream=True,
    stream_events=True,
)
for chunk in run_stream:
    pprint(dataclass_to_dict(chunk, exclude={"messages"}))
    print("---" * 20)

Usage

1. Create a virtual environment

   Open the Terminal and create a python virtual environment.

   Mac

   python3 -m venv .venv
   source .venv/bin/activate
   
2. Set your API key

   export OPENAI_API_KEY=xxx
   
3. Install libraries

   pip install -U openai rich agno
   
4. Run Agent

   Mac

   python cookbook/agent_concepts/multimodal/generate_image_with_intermediate_steps.py