Amazon Titan Image Generator v2
AWS is pleased to announce that the Amazon Titan Image Generator v2 model, which includes new features in Amazon Bedrock, is now generally available. Using reference photos, you may direct the development of images with Amazon Titan Image Generator v2, alter already-existing images, eliminate backdrops, create image variations, and safely alter the model to preserve subject consistency and brand style. This effective tool increases productivity, simplifies processes, and realises creative ideas.
In addition to all the capabilities of Amazon Titan Image Generator v1, Amazon Titan Image Generator v2 offers a variety of additional functions, such as:
- Image conditioning involves giving a user-supplied reference image and a text prompt; the outputs will then mimic the reference’s layout and structure.
- Colour palette advice for images – Give a list of hex codes together with the text prompt to fine-tune the colour scheme of the output images.
- Background removal: This feature automatically eliminates backgrounds from pictures that have several things in them.
- Subject consistency: Adjust the model so that the produced images maintain a particular subject (like a dog, shoe, or handbag).
Amazon Titan Image Generator v2 has new features
If you have never used an Amazon Titan model before, start by going to the Amazon Bedrock console and selecting Model access in the bottom left pane. Make a separate request for access to the Amazon Titan Image Generator G1 v2 in order to obtain access to the most recent Amazon Titan models.
The Amazon Titan Image Generator v2 in Amazon Bedrock specifications are as follows:
Conditioning of images
With the image conditioning option, you may intentionally and precisely sculpt your works. You can direct the model to focus on particular visual features, including edges, object outlines, and structural elements, or segmentation maps that identify separate regions and objects inside the reference image, by giving it a reference image (also known as a conditioning image).
Canny edge and segmentation are the two forms of image conditioning that AWS provide.
- The Amazon Titan Image Generator can then use this map, which was created by extracting the notable edges from the reference image using the Canny edge algorithm, to direct its generation. You can “draw” the basic outline of the image you want, and the model will work from there to add the finishing details, textures, and aesthetic.
- Even more precise control is offered via segmentation. You can direct the Amazon Titan Image Generator to produce material that corresponds with designated regions by providing the reference image, which allows you to designate particular objects or places within the image. Characters, objects, and other important elements can be precisely placed and rendered.
Use the Amazon Bedrock API, AWS SDK, or AWS Command Line Interface (AWS CLI) and select CANNY_EDGE or SEGMENTATION as the controlMode of textToImageParams with your reference picture to activate the image conditioning function.
The conditioning of colour
Most designers demand control over the colour palette in the created images because they want to produce images that adhere to colour branding rules.
You can create color-conditioned graphics using the Amazon Titan Image Generator v2 by providing a list of hex colours as inputs and using the colour palette to create images that follow colour branding rules. In order to generate an image using the supplied hex colours while inheriting the style from the reference image, you can optionally supply a reference image as input (optional).
The created graphic adheres to the brand’s colour rules by reflecting both the text prompt’s content and the designated colour scheme.
You can set taskType to COLOR_GUIDED_GENERATION with your prompt and hex codes to use the colour conditioning feature.
Background elimination
The ability to precisely and cleanly remove the background from an image is a crucial tool in the creative workflow, regardless of whether you’re trying to layer it over another scene or composite it onto a solid colour backdrop. With only one action, you may automatically erase the backdrop from your photographs. Using its clever detection and segmentation capabilities, Amazon Titan Image Generator v2 can effectively extract even intricate landscapes with overlapping elements.
An illustration of an iguana perched on a tree in a forest is displayed in the example. The iguana was recognised by the model as the primary object, and the woodland background was eliminated in favour of a translucent one. As a result, the iguana is able to stand out without the distracting surrounding foliage.
You can set taskType to BACKGROUND_REMOVAL with your input image to utilise the background removal capability.
Subject coherence while adjusting
Now, you may easily add particular subjects to scenes that are visually striking. Reference photos can be used to fine-tune the Amazon Titan model by teaching it about the distinctive features of the chosen subject, be it a product from the brand, the emblem of the company, or a cherished family pet.
After the model is optimised, all you have to do is give the Amazon Titan Generator a text prompt to produce photographs that preserve the subject’s naturalistic portrayal in a variety of creative scenarios. This creates a plethora of opportunities for visual storytelling, marketing, and advertising.
For instance, when fine-tuning, you could use an image with the description Ron the dog; during inference with the fine-tuned model, you could give the prompt as Ron the dog wearing a superhero cape, and you would receive a unique image in return.
Currently accessible
In the US East (North Virginia) and US West (Oregon) Regions, Amazon Bedrock is now selling the Amazon Titan Generator v2 model. See the complete list of regions for upcoming changes. Visit the Amazon Bedrock pricing page and the Titan product page to find out more.
