Nova Canvas - 背景移除
本节将移除图像的背景, 用透明像素替换它。
Amazon Nova Canvas模型的背景移除功能可以自动将图像的主体与背景隔离,返回一个具有8位透明度的PNG图像：
移除后：
将第一张图片保存为man-in-orange.png，保存到当前目录，运行代码：
from random import randint
import json
import base64
import boto3
client = boto3.client(service_name='bedrock-runtime', region_name="us-east-1")
source_image_path = "man-in-orange.png"
# Read image from disk.
with open(source_image_path, "rb") as image_file:
source_image_base64 = base64.b64encode(image_file.read()).decode("utf8")
inference_params = json.dumps({
"taskType": "BACKGROUND_REMOVAL",
"backgroundRemovalParams": {
"image": source_image_base64,
},
})
response = client.invoke_model(
body=inference_params,
modelId="amazon.nova-canvas-v1:0",
accept="application/json",
contentType="application/json",
)
response_body = json.loads(response.get("body").read()) # 读取并解析响应体
base64_image = response_body.get("images")[0] # 获取返回的 Base64 编码图片数据
# 将 Base64 编码的图片解码为二进制数据
base_64_image = base64.b64decode(base64_image)
# 将解码后的图片保存为文件
file_path = "1.png"
with open(file_path, "wb") as f:
f.write(base_64_image)
print(f"图片已成功保存为: {file_path}")
功能介绍
背景移除 - 从原始背景中分离对象，有助于创建多用途的营销素材。
请求示例
{
"taskType": "BACKGROUND_REMOVAL"，
"backgroundRemovalParams": {
"image": 字符串(Base64编码的图像)，
}
}
BACKGROUND_REMOVAL任务将返回一个具有全8位透明度的PNG图像，可以平滑、干净地隔离前景对象，并使其易于在图像编辑应用程序、演示文稿或网站中与其他元素合成。背景可以使用简单的自定义代码轻松更改为纯色。
注意: 背景移除任务会自动确定图像中哪些元素应被视为前景和背景。没有可以影响此的参数。
使用mask prompt删除对象
从图像中删除对象的一种方法是使用“maskPrompt”描述要删除的对象。下面的代码将使用mask prompt(花盆中的花)并从图像中删除所有花盆。
删除前：
删除后：
将第一张图片保存为three_pots.png, 放在代码目录，并运行：
from random import randint
import json
import base64
import boto3
client = boto3.client(service_name='bedrock-runtime', region_name="us-east-1")
source_image_path = "three_pots.jpg"
# Read image from disk.
with open(source_image_path, "rb") as image_file:
source_image_base64 = base64.b64encode(image_file.read()).decode("utf8")
inference_params = json.dumps({
"taskType": "INPAINTING",
"inPaintingParams": {
"image": source_image_base64,
"maskPrompt": "flowers in pots", # Description of the elements to remove
# We intentionally omit the "text" param. By doing so, the generated
# content will attempt to match the background.
},
"imageGenerationConfig": {
"numberOfImages": 1, # Number of variations to generate. 1 to 5.
"quality": "standard", # Allowed values are "standard" and "premium"
"cfgScale": 7.0, # How closely the prompt will be followed
"seed": randint(0, 858993459), # Use a random seed
},
})
response = client.invoke_model(
body=inference_params,
modelId="amazon.nova-canvas-v1:0",
accept="application/json",
contentType="application/json",
)
response_body = json.loads(response.get("body").read()) # 读取并解析响应体
base64_image = response_body.get("images")[0] # 获取返回的 Base64 编码图片数据
# 将 Base64 编码的图片解码为二进制数据
base_64_image = base64.b64decode(base64_image)
# 将解码后的图片保存为文件
file_path = "1.png"
with open(file_path, "wb") as f:
f.write(base_64_image)
print(f"图片已成功保存为: {file_path}")