MCP Media Generator

import boto3 import json import os import base64 import io from src.upload_image import upload_file_to_s3 # Import environment variables AWS_REGION = os.environ.get("AWS_REGION", "us-east-1") MODEL_ID = "amazon.nova-canvas-v1:0" AWS_ACCESS_KEY_ID = os.environ.get("AWS_ACCESS_KEY") AWS_SECRET_ACCESS_KEY = os.environ.get("AWS_SECRET_ACCESS_KEY") async def create_image(prompt,negative_prompt, quality, width, height, seed_value): """ Creates image using Amazon Nova Canvas model. :param prompt: Prompt for image. :param negative_prompt: What not to use in image. :param quality: Standard or premium quality :param width: Width of the picture :param height: Height of the picture. :param seed_value: Seed value for the picture :return: Base64 encoded image """ # Initiate Bedrock Client bedrock = boto3.client("bedrock-runtime", region_name=AWS_REGION, aws_access_key_id=AWS_ACCESS_KEY_ID, aws_secret_access_key=AWS_SECRET_ACCESS_KEY) # Set picture parameters model_input = json.dumps({ "taskType": "TEXT_IMAGE", "textToImageParams": { "text": prompt, "negativeText": negative_prompt }, "imageGenerationConfig": { "width": width, "height": height, "quality": quality, "cfgScale": 3, "seed": seed_value, "numberOfImages": 1 } }) # Invoke model response = bedrock.invoke_model( body=model_input, modelId=MODEL_ID, accept="application/json", contentType="application/json" ) # Read the response body response_body = json.loads(response.get("body").read()) # Extract the first image base64_image = response_body.get("images")[0] # Decode the Base64 string into bytes image_data = base64.b64decode(base64_image) # Create a file-like object from the bytes image_file = io.BytesIO(image_data) url = upload_file_to_s3(image_file) return f"The image URL is {url}. Display the image to user in response via markdown  syntax and also separately provide URL as link in markdown for download."