MAI Labs
Frequently asked questions
No questions found.
About MAI Labs
3 questions
MAI Labs is a limited public preview experience that allows users access to select MAI models for their own personal, non-commercial experimentation and research purposes. It is offered to support users' exploration and evaluation of, and submission of feedback relating to, models that are available in MAI Labs for limited public preview.
MAI Labs and the models available through it are experimental and may not work correctly. The models available in MAI Labs are AI and may make mistakes, including producing incorrect, incomplete, or unexpected results. Additionally, the models available in MAI Labs may not work in the same way the final version of the model will work.
MAI Labs is a limited public preview experience designed for early exploration and feedback on select MAI models for experimentation and research purposes. Unlike other AI experiences at Microsoft that are built for production use or integrated into finished products, MAI Labs is intentionally lightweight and experimental. It focuses on giving users access to models in limited public preview, with a limited feature set designed to showcase core capabilities rather than deliver a full product experience.
The models available in MAI Labs are AI and may make mistakes, including producing incorrect, incomplete, or unexpected results. Additionally, the models available in MAI Labs may not work in the same way the final version of the model will work.
Availability of MAI Labs varies by country. We're working to expand availability as regulatory and operational conditions allow.
Getting started and access
3 questions
Online at https://labs.microsoft.ai. You must be signed into your Microsoft Account to access MAI Labs. To sign into MAI Labs, follow these steps:
- Go to the MAI Labs website on your desktop or mobile device.
- Sign into your existing Microsoft Account by entering your credentials or creating a new account.
A Microsoft account is an account that you use to access Microsoft services, which can be a personal account or a Microsoft Entra ID (typically a work or school account).
You may only use MAI Labs for your own personal, non-commercial purposes to experiment with the limited public preview models and multi-model experiences available through MAI Labs. If you use a Microsoft Entra ID to access MAI Labs you must not use MAI Labs for work or school purposes.
MAI Labs is available to users over the age of 18. We may expand access to users under 18 subject to minimum age requirements that vary by country or region. https://support.microsoft.com/account-billing/list-of-countries-or-regions-that-have-statutory-age-limits-for-children-a7393f56-5107-44b1-be44-655589bdeb92
Using MAI Labs
3 questions
Once you're signed into MAI Labs at https://labs.microsoft.ai using your Microsoft Account, you're ready to start experimenting with the models that are available. Select a model from the model selection on the left-hand side of the page. In some cases, the model listed on the left-hand side of the page may represent a model family, so once you make your selection, you may have the option to experiment with different versions or options available in the model family by choosing the version/option in the drop-down menu. Once you've made your selections, submit a prompt. The output will vary depending on the type of model! The models in MAI Labs are in limited public preview and are experimental, so be mindful of the information you share and how much time you spend experimenting.
Currently, MAI Labs is available in the following languages:
- English
- Spanish
- French
- French (Canadian)
- German
- Italian
- Japanese
- Indonesian
Individual models & multi-model experiences available through MAI Labs may support additional languages.
You can explore MAI models in Foundry.
Models
3 questions
MAI Labs currently allows you to experiment with the following models:
Text Model
- MAI-Thinking-1: A chat-based text reasoning model – given a prompt, it produces an internal chain of thought before emitting a final response, allocating reasoning effort adaptively according to prompt complexity. It focuses on reasoning efficiency and agentic capability rather than a change in architecture or scale. See model card for more details.
Image Generation Models
- MAI-Image-2.6: Generates images, returning a single image per prompt. See model card and data summary for more details.
- MAI-Image-2.5: Generates and edits images, returning a single image per prompt. See model card and data summary for more details.
- MAI-Image-2.5-Flash: Generates and edits images, returning multiple images per prompt. See model card and data summary for more details.
- MAI-Image-2.5-Pro: Generates and edits images, returning multiple images per prompt. See model card for more details.
Transcription Models
- MAI-Transcribe-1.5: A transcription model that converts audio into written text. Audio clips can be uploaded to MAI Labs or recorded directly in the MAI Labs experience. See model card for more details.
Voice Generation Models
- MAI-Voice-2: A multilingual text-to-speech model that converts text prompts into a playable audio clip. The model supports different languages, voices and styles. See model card for more details.
- MAI-Voice-2-Flash: A multilingual text-to-speech model that converts text prompts into a playable audio clip. The model supports different languages, voices and styles. See model card for more details.
We may, but are not obligated to, make other models available for experimentation through MAI Labs from time to time. When we do, we will update these FAQs to provide more information on those models.
MAI Labs may include limited, experimental experiences that allow you to interact with multiple models at the same time. These types of experimental experiences may demonstrate how model capabilities may work when the models are combined, allow you to explore new ideas, or highlight our research. These experiences may range from interactive demos to research-driven explorations, and do not represent full product features or production use scenarios.
We may change, update, and add or remove experiences from MAI Labs at any time.
DuoAI is an experimental, real-time conversational voice experience available in MAI Labs that allows you to interact with two AI personas, Luna and Leo. DuoAI combines transcription, voice generation, and image generation models to demonstrate multi-model interaction in a single experience.
DuoAI is intended to be used in English. Performance cannot be guaranteed in other languages.
Data storage and deletion
7 questions
Images or other generated content will be saved for up to 29 days. After this time, images and other generated content are deleted from our system.
However, interactions with DuoAI are ephemeral and do not persist beyond a single session. This means that your DuoAI conversation, including images and other generated content during the DuoAI session, are not saved after that specific session ends, and you will not be able to access previous conversations in DuoAI.
You have two options:
- Delete a specific conversation: Hover over a conversation in the history panel on the left. Click the "X" that appears to remove that conversation.
- Delete all account data: Click your account name in the bottom-left corner, then select "Data and Privacy." Follow the prompts to delete your history.
You will not be able to delete specific DuoAI conversations or content generated through use of DuoAI since DuoAI conversations are not saved after the specific session ends.
When you experiment with a model, or a multi-model experience, in MAI Labs, we may use your prompts and the generated responses or creations to operate and improve MAI Labs. We do not currently train the models or any multi-model experience available through MAI Labs using your prompts or responses or creations generated by the models based on your prompts.
To learn more about Microsoft's commitment to privacy, visit the Microsoft Privacy Statement and Privacy at Microsoft.
If you share a file with MAI Labs (for example, uploading an image for image editing or uploading an audio file for transcription), the file will be stored for up to 29 days and then automatically deleted. The file will be treated just like any other prompts or generated content, which you can delete at any time unless you upload it during a DuoAI session, in which case the file does not persist after the session ends.
MAI Labs may ask for your permission to process biometric data—like faces or hands—when you use features such as image upload to edit images. Consent for biometric processing applies anytime you upload an image while using MAI Labs. This includes through your use of image generation and editing tools in models available in MAI Labs. You only need to give consent once, and it applies across all MAI Labs experiences that involve image input.
Biometric data is only processed if you upload it, and it's used solely to respond to your request. For example, if you snap a selfie and ask a model in MAI Labs for help choosing a new hairstyle, it may analyze your facial features to help craft a response.
You're always in control of what you share. If you prefer not to have biometric data processed, simply avoid using these features. Do not share images of others without their consent.
You shouldn't provide any confidential information or sensitive personal data that you would not want Microsoft to use for purposes explained in this FAQ and in the Microsoft Privacy Statement. This includes data that might reveal, for example, your race, religion, sexual orientation, or health status.
At any time, you can delete your prompts and generated content from MAI Labs, except for prompts and generated content in DuoAI, which do not persist after the applicable DuoAI session ends.
Responsible AI and safety
3 questions
At Microsoft, we take our commitment to responsible AI seriously. The models that are made available to you on MAI Labs are developed in accordance with our AI principles, which demonstrate commitment to making sure AI systems are developed responsibly and in ways that warrant people's trust. We've designed the MAI Labs user experience to keep humans at the center and developed safeguards to minimize errors and avoid misuse, and we are continually improving the experience. To learn more about how to use MAI Labs responsibly, please review the MAI Labs Terms of Use.
While we have designed the models available in MAI Labs to avoid generating unexpected or offensive content, the models are experimental and you may still see unexpected results. We're continuously working to improve our technology to prevent harmful content.
If you encounter harmful or inappropriate content, please report this to Microsoft using MAI Labs' Report a Concern form.
Repeated violations of our Code of Conduct in the MAI Labs Terms of Use may result in automatic temporary suspension of your use of MAI Labs. Multiple suspensions may lead to permanent restriction of your use of MAI Labs. Permanent suspension may be appealed by submitting a request through the MAI Labs Report a Concern form.