This AI Agent was built to address two of the main problems when a user tries to generate an image with AI:
-
Generating the right prompt to get what the user wants
-
Generate more than one image per user request (if asked to), which can also be of different topics each.
Is capable of generate different images with different styles, but is mainly focus on generate realistic images.
This agent is deployed in Google Cloud Platform, so you can test it by just register for free here - The images, once generated, will be deleted after 7 days.
The AI Agent was built using the PydanticAI agent framework, and some of the Gemini models, mainly Gemini 2.5 Pro and Imagen 3.
The Cloud infrastructure was developed using Terraform and deployed through CICD pipelines with CloudBuild.
The app was containerized using Docker and deployed on two serverless instances on CloudRun:
-
One related to the API that contains all the logic of the AI Agent (built with FastAPI and secured using JWT).
-
Other related to the frontend section (built with Streamlit).
All the chat sessions data are stored on BigQuery, whereas the generated images are stored in Google Cloud Storage buckets - which are automatically deleted after 7 days.