Amazon Titan Picture Generator and watermark detection API at the moment are accessible in Amazon Bedrock

Throughout AWS re:Invent 2023, we introduced the preview of Amazon Titan Picture Generator, a generative synthetic intelligence (generative AI) basis mannequin (FM) that you should use to rapidly create and refine real looking, studio-quality photographs utilizing English pure language prompts.

I’m joyful to share that Amazon Titan Picture Generator is now usually accessible in Amazon Bedrock, supplying you with a simple technique to construct and scale generative AI functions with new picture technology and picture modifying capabilities, together with immediate customization of photographs.

In my earlier submit, I additionally talked about that every one photographs generated by Titan Picture Generator include an invisible watermark, by default, which is designed to assist scale back the unfold of misinformation by offering a mechanism to establish AI-generated photographs.

I’m excited to announce that watermark detection for Titan Picture Generator is now usually accessible within the Amazon Bedrock console. Right now, we’re additionally introducing a brand new DetectGeneratedContent API (preview) in Amazon Bedrock that checks for the existence of this watermark and helps you verify whether or not a picture was generated by Titan Picture Generator.

Let me present you the right way to get began with these new capabilities.

Prompt picture customization utilizing Amazon Titan Picture Generator
Now you can generate new photographs of a topic by offering as much as 5 reference photographs. You possibly can create the topic in numerous scenes whereas preserving its key options, switch the type from the reference photographs to new photographs, or combine kinds from a number of reference photographs. All this may be completed with out extra immediate engineering or fine-tuning of the mannequin.

For this demo, I immediate Titan Picture Generator to create a picture of a “parrot consuming a banana.” Within the first try, I exploit Titan Picture Generator to create this new picture with out offering a reference picture.

Observe: Within the following code examples, I’ll use the AWS SDK for Python (Boto3) to work together with Amazon Bedrock. Yow will discover extra code examples for C#/.NET, Go, Java, and PHP within the Bedrock Person Information.

import boto3
import json

bedrock_runtime = boto3.consumer(service_name="bedrock-runtime")

physique = json.dumps(
    {
        "taskType": "TEXT_IMAGE",
        "textToImageParams": {
            "textual content": "parrot consuming a banana",   
        },
        "imageGenerationConfig": {
            "numberOfImages": 1,   
            "high quality": "premium", 
            "peak": 768,
            "width": 1280,
            "cfgScale": 10, 
            "seed": 42
        }
    }
)
response = bedrock_runtime.invoke_model(
    physique=physique, 
    modelId="amazon.titan-image-generator-v1",
    settle for="software/json", 
    contentType="software/json"
)

You possibly can show the generated picture utilizing the next code.

import io
import base64
from PIL import Picture

response_body = json.hundreds(response.get("physique").learn())

photographs = [
    Image.open(io.BytesIO(base64.b64decode(base64_image)))
    for base64_image in response_body.get("images")
]

for img in photographs:
    show(img)

Right here’s the generated picture:

Then, I exploit the brand new immediate picture customization functionality with the identical immediate, however now additionally offering the next two reference photographs. For simpler comparability, I’ve resized the photographs, added a caption, and plotted them aspect by aspect.

Right here’s the code. The brand new immediate customization is out there by the IMAGE_VARIATION process:

# Import reference photographs
image_path_1 = "parrot-cartoon.png"
image_path_2 = "bird-sketch.png"

with open(image_path_1, "rb") as image_file:
    input_image_1 = base64.b64encode(image_file.learn()).decode("utf8")

with open(image_path_2, "rb") as image_file:
    input_image_2 = base64.b64encode(image_file.learn()).decode("utf8")

# ImageVariationParams choices:
#   textual content: Immediate to information the mannequin on the right way to generate variations
#   photographs: Base64 string illustration of a reference picture, as much as 5 photographs are supported
#   similarityStrength: Parameter you possibly can tune to manage similarity with reference picture(s)

physique = json.dumps(
    {
        "taskType": "IMAGE_VARIATION",
        "imageVariationParams": {
            "textual content": "parrot consuming a banana",  # Required
            "photographs": [input_image_1, input_image_2],  # Required 1 to five photographs
            "similarityStrength": 0.7,  # Vary: 0.2 to 1.0
        },
        "imageGenerationConfig": {
            "numberOfImages": 1,
            "high quality": "premium",
            "peak": 768,
            "width": 1280,
            "cfgScale": 10,
            "seed": 42
        }
    }
)

response = bedrock_runtime.invoke_model(
    physique=physique, 
    modelId="amazon.titan-image-generator-v1",
    settle for="software/json", 
    contentType="software/json"
)

As soon as once more, I’ve resized the generated picture, added a caption, and plotted it aspect by aspect with the initially generated picture.

You possibly can see how the parrot within the second picture that has been generated utilizing the moment picture customization functionality resembles in type the mixture of the offered reference photographs.

Watermark detection for Amazon Titan Picture Generator
All Amazon Titan FMs are constructed with accountable AI in thoughts. They detect and take away dangerous content material from information, reject inappropriate person inputs, and filter mannequin outputs. As content material creators create realistic-looking photographs with AI, it’s essential to advertise accountable improvement of this expertise and scale back the unfold of misinformation. That’s why all photographs generated by Titan Picture Generator include an invisible watermark, by default. Watermark detection is an revolutionary expertise, and Amazon Internet Companies (AWS) is among the many first main cloud suppliers to broadly launch built-in watermarks for AI picture outputs.

Titan Picture Generator’s new watermark detection characteristic is a mechanism that lets you establish photographs generated by Amazon Titan. These watermarks are designed to be tamper-resistant, serving to enhance transparency round AI-generated content material as these capabilities proceed to advance.

Watermark detection utilizing the console
Watermark detection is usually accessible within the Amazon Bedrock console. You possibly can add a picture to detect watermarks embedded in photographs created by Titan Picture Generator, together with these generated by the bottom mannequin and any custom-made variations. In case you add a picture that was not created by Titan Picture Generator, then the mannequin will point out {that a} watermark has not been detected.

The watermark detection characteristic additionally comes with a confidence rating. The arrogance rating represents the arrogance degree in watermark detection. In some instances, the detection confidence could also be low if the unique picture has been modified. This new functionality permits content material creators, information organizations, threat analysts, fraud detection groups, and others to higher establish and mitigate deceptive AI-generated content material, selling transparency and accountable AI deployment throughout organizations.

Watermark detection utilizing the API (preview)
Along with watermark detection utilizing the console, we’re introducing a brand new DetectGeneratedContent API (preview) in Amazon Bedrock that checks for the existence of this watermark and helps you verify whether or not a picture was generated by Titan Picture Generator. Let’s see how this works.

For this demo, let’s verify if the picture of the inexperienced iguana I confirmed within the Titan Picture Generator preview submit was certainly generated by the mannequin.

I outline the imports, arrange the Amazon Bedrock boto3 runtime consumer, and base64-encode the picture. Then, I name the DetectGeneratedContent API by specifying the inspiration mannequin and offering the encoded picture.

import boto3
import json
import base64

bedrock_runtime = boto3.consumer(service_name="bedrock-runtime")

image_path = "green-iguana.png"

with open(image_path, "rb") as image_file:
    input_image_iguana = image_file.learn()

response = bedrock_runtime.detect_generated_content(
    foundationModelId = "amazon.titan-image-generator-v1",
    content material = {
        "imageContent": { "bytes": input_image_iguana }
    }
)

Let’s verify the response.

response.get("detectionResult")
'GENERATED'
response.get("confidenceLevel")
'HIGH'

The response GENERATED with the arrogance degree HIGH confirms that Amazon Bedrock detected a watermark generated by Titan Picture Generator.

Now, let’s verify one other picture I generated utilizing Secure Diffusion XL 1.0 on Amazon Bedrock. On this case, a “meerkat dealing with the sundown.”

I name the API once more, this time with the picture of the meerkat.

image_path = "meerkat.png"

with open(image_path, "rb") as image_file:
    input_image_meerkat = image_file.learn()

response = bedrock_runtime.detect_generated_content(
    foundationModelId = "amazon.titan-image-generator-v1",
    content material = {
        "imageContent": { "bytes": input_image_meerkat }
    }
)

response.get("detectionResult")
'NOT_GENERATED'

And certainly, the response NOT_GENERATED tells me that there was no watermark by Titan Picture Generator detected, and due to this fact, the picture almost certainly wasn’t generated by the mannequin.

Utilizing Amazon Titan Picture Generator and watermark detection within the console
Right here’s a brief demo of the right way to get began with Titan Picture Generator and the brand new watermark detection characteristic within the Amazon Bedrock console, put collectively by my colleague Nirbhay Agarwal.

Availability
Amazon Titan Picture Generator, the brand new immediate customization capabilities, and watermark detection within the Amazon Bedrock console can be found at this time within the AWS Areas US East (N. Virginia) and US West (Oregon). Verify the full Area listing for future updates. The brand new DetectGeneratedContent API in Amazon Bedrock is out there at this time in public preview within the AWS Areas US East (N. Virginia) and US West (Oregon).

Amazon Titan Picture Generator, now additionally accessible in PartyRock
Titan Picture Generator is now additionally accessible in PartyRock, an Amazon Bedrock playground. PartyRock offers you a no-code, AI-powered app-building expertise that doesn’t require a bank card. You should utilize PartyRock to create apps that generate photographs in seconds by deciding on out of your alternative of picture technology fashions from Stability AI and Amazon.

Extra assets
To be taught extra concerning the Amazon Titan household of fashions, go to the Amazon Titan product web page. For pricing particulars, verify Amazon Bedrock Pricing.

Give Amazon Titan Picture Generator a attempt in PartyRock or discover the mannequin’s superior picture technology and modifying capabilities within the Amazon Bedrock console. Ship suggestions to AWS re:Put up for Amazon Bedrock or by your common AWS contacts.

For extra deep-dive technical content material and to have interaction with the generative AI Builder group, go to our generative AI house at group.aws.

— Antje