Microsoft Photos Restyle Image and Image Creator: Responsible AI FAQ
Restyle Image and Image Creator are new features in the Microsoft Photos app, exclusive on Copilot+ PCs, that let you reimagine your photos or create new imagery with the help of AI. Along with giving you a brief overview of Restyle Image and Image Creator, this article will provide the following information:
-
What are the intended uses for Restyle Image and Image Creator?
-
How were these features evaluated? What metrics are used to measure performance?
With Restyle Image, you can add original, contextual AI-generated styles to your photos, by either choosing one of our curated, pre-built style options and adjusting the creativity level, or by creating your own custom style with a text description.
To use Restyle Image:
-
Open up any of your photos with the Microsoft Photos app.
-
At the top of the Photos app, select Edit image or use the Ctrl + E keyboard shortcut.
-
Select the Restyle Image option.
-
You can select one of the preset prompts, or enter a custom description. You can also adjust the creativity level.
With Image Creator, you can create new, unique images simply by entering a text description of what you’d like to see.
To use Image Creator:
-
Open the Microsoft Photos app and select the Image Creator option in the navigation panel on the left.
-
Enter a text prompt into the box. The Photos app will start creating images based on your description such as floral arrangement.
-
Continue to scroll down the page to keep generating new images.
-
Once you’ve found an image you like, you can save it or use it to generate other similar images.
What can Restyle Image and Image Creator do?
Both features use proprietary and open-source AI models that are finetuned by Microsoft. The models come with your Copilot+ PC or can be automatically downloaded from Microsoft Store the first time you try these features. Image Creator uses a Text-to-Image model to turn the text you’ve entered into a series of images based on this description. Restyle Image uses a Sketch-to-Image model, using the text prompt you entered or selected to create the style and apply it to your selected photo.
In Restyle Image, if the photo you selected includes prominent faces in the foreground, we want to preserve the facial and human characteristics in the photo. To do this, we use another AI model (also on the device) to detect if there are faces in the photo. If faces are found, then we only apply the style to the background of the photo. If no faces are found, the style is applied to the entire photo. Note that no identifiable, biometric data is collected, processed, or stored for using this feature – we are only using the AI model to determine if there may be faces in the photo, and thus preserve them from distortion.
What are the intended uses for Restyle Image and Image Creator?
Restyle Image lets you transform any photo with a single click. Use it to turn a landscape shot into artwork using styles like Watercolor or Impressionism, or create a fun background filled with balloons for your birthday party photos.
Image Creator helps you turn abstract concepts and ideas into original, unique images that you can pull inspirations from. Enter the prompt describing what you’d like to see, like book cover ideas or event venue decorations, and let AI generate the visuals for you.
How were these features evaluated? What metrics are used to measure performance?
To ensure quality in AI-generated content, we ran rigorous testing on these features, both quantitative and qualitative, to help ensure that the models create appropriate content in response to the user-entered or selected prompts. In particular, we ran tests to identify cases where potentially harmful images could be used or created, and we have improved our content moderation mechanisms based on the results of those tests.
We will also continuously monitor and evaluate the performance of our features through a number of user engagement methods, including user interviews, reviews of usage metrics, and integrated product feedback mechanisms. Several key metrics are used to measure our quality and performance, including user engagement rates, feedback volume, and other usability metrics.
What are the limitations of Restyle Image and Image Creator? How can users minimize the impact of the limitations when using the system?
Currently, both Restyle Image and Image Creator are optimized for prompts in US English. Prompts entered in other languages will be translated and results may vary due to the translation.
To use Restyle Image and Image Creator, you must be connected to the internet and signed into your Microsoft account, in order to access cloud services that help ensure safe use of AI. We currently do not support these features for Microsoft Entra account users.
Commitment to responsible AI and privacy
We take our commitment to responsible AI seriously.
All users must comply with the Code of Conduct in the Microsoft Services Agreement.
We have put controls in place to help prevent the generation of harmful, offensive, or inappropriate images that violate our Code of Conduct. User prompts undergo abuse monitoring via internal content safety mechanisms. When our system detects that a potentially harmful image could be generated, it automatically blocks the operation and informs the user. Abuse of these features, such as repeated attempts to create harmful content or other violations of the Code of Conduct, may result in automatic suspension of your use of the Restyle Image and Image Creator features.
To allow us to effectively investigate and remediate cases where harmful content violating our responsible AI principles is being generated, when you use Restyle Image and Image Creator, we collect data around the prompts you enter and how the Photos app and models have responded, in accordance to our privacy statement.
For all images, we have implemented content credentials based on the C2PA standard to help identify that the images are AI edited or generated.