BRIA Attribution Model

This repository contains a Jupyter notebook demonstrating how to integrate the Bria attribution service providing full legal indemnity when hosting Bria's models on your cloud / environment.

There are 2 parts to this integration:

1. Installing the Bria attribution model for calculating embeddings (irreversible vectors) of your generated images.
2. API Integration for sharing the embeddings with Bria.

Overview

The code_example.ipynb notebook provides a complete workflow for:

1. Setting up and running the BRIA attribution model using Docker and Triton inference server
2. Generating image embeddings using the BRIA model
3. Sending attribution API requests with the generated embeddings

Prerequisites

Before running this example, ensure you have the following installed:

• Docker
• Python 3 with the packages (pip install -r ./requirements.txt --use-deprecated=legacy-resolver)
• NVIDIA GPU with CUDA support (recommended)

Setup Instructions

1. API Token Configuration

You'll need to configure your BRIA API token to access the model repository and attribution service. Replace the placeholder values in the notebook:

api_token = '<API_TOKEN>'

Required substitutions:

• <API_TOKEN>: Your BRIA API token for accessing the attribution service

2. Model Deployment

The notebook will automatically:

• Fetch ECR credentials using your API token from the BRIA attribution service
• Login to Amazon ECR using the fetched credentials
• Pull the BRIA attribution model Docker image
• Start the Triton inference server with the model

The model will be available on:

• Port 8000: HTTP endpoint
• Port 8001: Metrics endpoint
• Port 8002: GRPC endpoint

3. Client Configuration

Required Substitutions in the Notebook:

1. BRIA Embeddings URL:

   url = '<BRIA_EMBEDDINGS_URL>:8000'

   Replace <BRIA_EMBEDDINGS_URL> with your server's IP address or hostname (localhost if running on the same machine).

2. Image Path:

   image_path = '<IMAGE_PATH>'

   Replace <IMAGE_PATH> with the path to your input image file.

3. API Token (in attribution request):

   api_token = '<API_TOKEN>'

   Replace <API_TOKEN> with your BRIA API token.

Usage

Step 1: Run the Docker Setup Cell

Execute the first cell to:

• Configure your API token
• Fetch ECR credentials from the BRIA attribution service
• Login to ECR using the fetched credentials
• Pull and run the BRIA attribution model container

Step 2: Install Dependencies

Run the installation cell to import required libraries:

• requests for API calls
• tritonclient.http for Triton inference
• PIL for image processing
• embedder module (BRIAEmbedder class)

Step 3: Generate Embeddings

Execute the inference request cell to:

• Connect to the Triton server
• Load and process your image
• Generate embeddings using the BRIA model

Step 4: Send Attribution Request

Run the attribution API cell to:

• Create a unique embedding ID
• Send the embeddings to the attribution service
• Handle the API response

Configuration Options

Model Parameters

The notebook uses these default parameters:

• Model Version: 2.3
• Model Name: 'replace-backgroud'
• Agent: Your API token

API Request Format

The attribution API expects the following JSON structure:

{
  "embeddings_base64": "<base64_encoded_embeddings>",
  "embeddings_uid": "<unique_identifier>",
  "model_version": 2.3,
  "agent": "<API_TOKEN>",
  "model_name": "replace-backgroud"
}