Amazon's Titan AI Image Generator Is Here

Tech giant Amazon has joined the AI image generation space with the introduction of its Titan Image Generator on the Amazon Web Services (AWS) development platform, Bedrock.

During his keynote speech at the AWS re:Invent 2023 event, Swami Sivasubramanian, AWS vice president of databases, analytics, and machine learning, previewed Titan Image Generator.

What is Amazon Titan?

Titan Image Generator is a new AI service from AWS that can generate high-quality, realistic images from text prompts or image inputs. It includes built-in safeguards against generating toxic, biased, or incorrect content.

The service supports prompting and image generation in English, with maximum input text lengths of 77 tokens and input image sizes of 25MB. Titan Image Generator aims to make AI-generated image creation more accessible and practical for business applications.

Amazon Titan Capabilities

Titan Image Generator offers a range of image editing capabilities:

• Text to image generation: Generate high-quality, realistic images using natural-language prompts.
• Image customization: Securely customize the model using company data to produce images consistent with your brand style.
• Image editing: Edit an existing or generated image with a text prompt. No image mask needed.
• Image variations: Edit parts of an image to remove or replace certain objects with a mask area you define.
• Image outpainting: Replace an existing background to generate lifestyle images with a mask area you define.
• Image Re-sizing: Extend an image’s borders with additional details while retaining the main subject of the image.

How To Setup Amazon Titan Image Generator

The image generator model is part of Amazon’s Bedrock developer platform. To try Titan, first create an Amazon Web Services console account.

Once logged in, open the “All Services” page and locate the “Amazon Bedrock” service.

Next, open the “Image Generation” tab under the Playgrounds menu.

Click “Select Model” and choose the “Titan Image Generator G1” model.

You can now start typing an image description to generate images.

Image Generation Examples

Here is an example image generated with the prompt “green iguana”:

You can make edits to the generated image, like creating variations, editing portions, or using it to generate other images.

To demonstrate the “edit” capability, I modified the prompt to “blue iguana” and obtained this result:

The result looks really cool and the details are very impressive.

You can adjust the configurations to modify how the output looks:

• Mode: The model generates a new image, or edits an existing image, based on the image you supply in Reference image.
• Negative prompt: Items or concepts that you don’t want the model to generate, such as cartoon or violence.
• Reference image: The image on which to generate the response or that you want the model to edit.
• Orientation: Choose between portrait or landscape mode.
• Number of images: You adjust the number of output images between 1 to 5.
• Advanced configurations: The inference parameters to pass to the model. Prompt strength is how strong the prompt influence the output image. Seed is a series of numbers that tells the AI how to generate an image.

Here is an example using the image editing feature to modify part of an image:

Prompt: 3d render of dinosour character, style of clay animation, standing on top of a rock surrounded by water, style of chibi, simple, minimal, no fine details, clay material, art toy, soft sculpture

In this example, I added the phrase “standing on top of a rock surrounded by water” on the original prompt. Here’s the final result:

How Amazon Titan Compares to Dall-E 3 and Midjourney

In order to see how Titan compares to two of the most popular and powerful AI image generators, I generated images with a common prompt and put them side by side.

Prompt: a straight on profile image of a gold skull sat on a black marble background, realistic metallic gold texture, high definition, ultra realistic

Looking at these results, I could say that Titan did an impressive job with realism compared to the other two. The reflection on the table looks very accurate.

Here’s another one:

Prompt: 3d render of dinosour character, style of clay animation, style of chibi, simple, minimal, no fine details, clay material, art toy, soft sculpture

While all outputs look cute and squishy, I prefer the Midjourney image for its more vibrant colors and cuter expression.

Pricing

Titan Image Generator offers flexible pricing to suit different needs and budgets, whether generating standard mockup images or premium marketing materials:

Remember, Titan is currently in preview so prices could change before general availability. Visit the official Bedrock website for the most up-to-date details.

Final Thoughts

In conclusion, Amazon’s entry into the AI image generation landscape with Titan Image Generator marks a welcome addition to its developer platform.

The impressive quality of images generated in the preview version rivals those of established competitors like OpenAI’s Dall-E 3 and Midjourney. Coupled with highly competitive pricing, Titan offers a compelling choice for developers seeking an image generation service for their applications.

Anticipation is high for the official public release of the API service. The possibilities for building innovative applications powered by Titan are vast, and I personally can’t wait to see what the developer community creates.