ガーデニングの新時代 ! Amazon Bedrock で理想の庭を実現してみた ~ GreenSnap 株式会社による生成 AI 実装解説 ~

import boto3 
import json 
import base64

# 理想の庭画像のファイルパス
INPUT_FILE_PATH ="{path/to/input/image}"

bedrock_runtime = boto3.client(
    service_name="bedrock-runtime",
    region_name="us-east-1"
)

def call_claude_haiku(base64_string):
    prompt_config = {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 4096,
        "messages": [
            {
                "role": "user",
                "content": [
                    {
                        "type": "image",
                        "source": {
                            "type": "base64",
                            "media_type": "image/png",
                            "data": base64_string,
                        },
                    },
                    # 庭の画像を説明してもらうプロンプト
                    {"type": "text", "text": "これは庭の画像です。庭の部分に焦点をあててこの庭の光景が伝わるように英語で500byte以内で教えて下さい"},
                ],
            }
        ],
    }
    body = json.dumps(prompt_config)
    model_id = "anthropic.claude-3-haiku-20240307-v1:0"
    accept = "application/json"
    content_type = "application/json"

    response = bedrock_runtime.invoke_model(
        body=body, modelId=model_id, accept=accept, contentType=content_type
    )
    return json.loads(response.get("body").read())

if __name__ == "__main__":
    with open(INPUT_FILE_PATH, "rb") as image_file:
        input_image = base64.b64encode(image_file.read()).decode("utf8")
    bed_rock_response = call_claude_haiku(input_image)
    result_prompt = bed_rock_response.get("content")[0].get("text")
    print(result_prompt) # 生成された理想の庭のprompt

import cv2

# 画像のパス
input_image_path = "{input/image/path}"
# マスク画像の保存先(拡張子はpngにする)
output_mask_image_path = "{output/image/path}"

if __name__ == "__main__":
    img = cv2.imread(input_image_path)
    roi = cv2.selectROI("マスク領域選択", img)
    x1, y1, w, h = roi
    x2, y2 = x1 + w, y1 + h
    img[:] = 255
    img[y1:y2, x1:x2] = 0
    cv2.imwrite(output_mask_image_path, img)

import boto3
import json
import base64
from PIL import Image
from io import BytesIO

if __name__ == "__main__":
    # 庭画像のパス
    input_image_path = "images/garden/target.png"
    # 生成したマスク画像のパス
    mask_image_path = "images/garden/mask_image.png"
    # アレンジ結果を保存するファイルパス
    output_image_path = "images/garden/garden_output.png"
    # 生成したプロンプト
    prompt = "This image depicts a stunning Japanese-style garden. The scene is filled with lush, verdant foliage, including towering trees with vibrant green leaves that create a canopy overhead. A winding stream flows through the garden, its banks lined with mossy rocks and gravel pathways. In the center, a traditional wooden gate or torii arch stands, adding an elegant cultural element to the landscape."

    with open(input_image_path, "rb") as image_file:
        input_image_base64 = base64.b64encode(image_file.read()).decode("utf8")
    with open(mask_image_path, "rb") as image_file:
        mask_image_base64 = base64.b64encode(image_file.read()).decode("utf8")

    bedrock_runtime = boto3.client(service_name="bedrock-runtime", region_name="us-east-1")
    body = json.dumps({
        "taskType": "INPAINTING",
        "inPaintingParams": {
            "text": prompt,
            "image": input_image_base64,
            "maskImage": mask_image_base64,
        },
        "imageGenerationConfig": {
            "numberOfImages": 1,  # Range: 1 to 5
            "quality": "premium",  # Options: standard or premium
            "cfgScale": 7.5,  # Range: 1.0 (exclusive) to 10.0
            "seed": 42  # Range: 0 to 214783647
        }
    })
    response = bedrock_runtime.invoke_model(
        body=body,
        modelId="amazon.titan-image-generator-v1",
        accept="application/json",
        contentType="application/json",
    )
    response_body = json.loads(response.get("body").read())
    images = [Image.open(BytesIO(base64.b64decode(base64_image))) for base64_image in response_body.get("images")]
    for idx, img in enumerate(images):
        img.save(output_image_path, quality=100)