MAI Image 2.7 is Microsoft's image generation and editing model family; Pexo offers a separate image-to-video workflow. The right choice depends on the output: a refined still image, an interactive editing session, or a finished video. Microsoft introduced MAI Image 2.7, Flash, and Canvas on November 17, 2026, but the model, workspace, and delivery process are different parts of that announcement.
What does MAI Image 2.7 do?
Microsoft's model card describes MAI Image 2.7 as a diffusion-based system for text-to-image generation and image-to-image editing. Text can describe a new image, while image inputs provide material for revisions. Its output is an image, so a buyer should not interpret an image-model launch as a native video-model launch. Microsoft model card.
The practical distinction is between creating and revising. A creative team might generate an initial composition, then request a specific object change while retaining the rest of the scene. These requests need different review criteria. A plausible new composition may be acceptable during exploration; a revision to an approved product picture needs much closer attention to preserved details.
Think of an image model as one component in a production process. The brief still determines what should appear, the reference establishes important visual facts, and a reviewer decides whether the result is usable. A model's ability to follow an instruction does not remove the need to check the actual output against the intended placement.
| Name | What it refers to | What to evaluate |
|---|---|---|
| Pexo | A separate creative production workflow | Whether approved images become a suitable finished video |
| MAI Image 2.7 | An image generation and editing model | Quality and control for the intended still image |
| MAI Image 2.7 Flash | The efficiency-oriented model variant | Whether the result meets the brief at the desired throughput |
| Canvas | An interactive workspace powered by Flash | Whether its editing interaction fits the team |
How is Flash different from the standard model?
Microsoft positions Flash for faster, lower-cost production, while the standard MAI Image 2.7 emphasizes fidelity and editing control. The launch announcement gives approximate public-preview prices of $35 per 1,000 images for the standard model and $10 for Flash. These are dated launch figures, not a quote for every provider or workflow. Confirm the applicable service price before estimating a project. Microsoft's launch announcement.
A useful comparison measures accepted results rather than initial generations. If one variant produces more usable compositions from the same brief, its total project cost may differ from the simple per-image price. Record rejected attempts and reviewer effort alongside generation charges. This is an evaluation method, not a claim about which variant wins your particular workload.
Separate exploration from final approval. A team may tolerate rough details while deciding the visual direction but require exact product identity in a delivery asset. Test both phases with the material the team actually uses. An attractive demonstration image cannot establish how the model handles a difficult reflective label or an unusually shaped product.
What is Canvas?
Canvas is the interactive image workspace announced inside MAI Labs, powered by MAI Image 2.7 Flash. Microsoft describes a workflow beginning with prompts or sketches, followed by selecting image regions to change. This makes Canvas an interface for working with the model rather than a third, independently named image model. Canvas in the launch announcement.
The workspace question is about how a person prefers to revise an image. Some tasks benefit from pointing at an exact area and describing the desired replacement. Other tasks are better organized as repeated requests in an application. Before switching tools, identify where the current workflow loses time: describing the edit, reviewing it, organizing variations, or exporting the accepted asset.
Preserve an approved baseline before experimenting. A revision that improves the background might change a detail that was already correct. Comparing each new result against the accepted baseline is more reliable than comparing only against the immediately preceding experiment. Keep the brief, source references, and selected output together so another reviewer can understand the decision.
Where can you access the models?
Microsoft Foundry lists MAI-Image-2.7 and MAI-Image-2.7-Flash in preview, with generation and editing APIs. Its documentation describes prerequisites including an Azure subscription, an appropriate Foundry project, and resource permissions. API access therefore involves a deployment and application workflow, not just visiting a consumer image page. Microsoft Foundry documentation.
The public MAI Labs page has a different context. Its onboarding notice describes MAI Labs as a limited preview for personal, non-commercial use, and says it should not be used for school or work activities. Do not assume that terms from a production service automatically apply to the experimental interface. Check the terms for the exact access path you intend to use. MAI Labs product page.
| Access question | Why it matters | What to record |
|---|---|---|
| Which service are you using? | Interfaces can have different terms | The actual service and account |
| Is the model a preview? | Availability and behavior can change | Model version and documentation date |
| Who can manage the deployment? | Application setup requires appropriate access | The responsible project owner |
| How will outputs be reviewed? | API success is not creative approval | An explicit acceptance process |
What should you check in an edited image?
Start with the requested change, then inspect everything that should have remained stable. A successful object recolor should not silently alter an adjacent logo or move a background feature. The relevant comparison is the source against the generated result, viewed at the size and crop used in the final placement. A thumbnail can conceal problems that become obvious in a close-up.
Text deserves a separate review. Check spelling, letter shapes, punctuation, and whether the supplied wording remains intact. For a branded asset, also compare the mark against the approved reference. An image that looks convincing at a glance may still contain an unusable detail. Treat this as an ordinary creative quality check rather than a promise that any model always preserves text.
Transparency also needs inspection in context. Microsoft documents transparent-background support, but the production test is whether the delivered edge works on the intended background. Review fine outlines, small gaps, and semi-transparent regions. A clean-looking asset on white may behave differently when placed over a dark layout or animated sequence.
| Review area | Compare against | Common acceptance question |
|---|---|---|
| Requested edit | The written change request | Did the intended change happen? |
| Preserved content | The approved source | Did unrelated details remain acceptable? |
| Text and branding | Exact supplied copy and marks | Is every visible detail correct? |
| Final placement | The real crop and background | Does the image work where it will be used? |
Does MAI Image 2.7 make videos?
The documented MAI Image 2.7 model output is still imagery. A sequence of separately generated images does not by itself establish a supported video-generation workflow. If your brief asks for movement, editing, sound, or a complete story, identify the tool responsible for those parts rather than assuming they are included in an image service.
Pexo's relevant role is the path from visual ideas or approved images to a finished video. Its public materials describe a conversational production workflow; its image studio is also relevant when creating source visuals. This article does not establish that Pexo includes MAI Image 2.7, nor does it claim that Pexo is Microsoft's Canvas workspace. Pexo, Pexo skills.
For a product campaign, the still image and the video have different approval questions. The still may need a specific composition, while the video also needs understandable pacing and a clear ending. Approve the visual source before treating it as the basis for further production. This gives the video review a stable reference instead of a moving target.
| Desired deliverable | Main decision | Completion means |
|---|---|---|
| A single generated image | Choose an image workflow | The accepted still matches the brief |
| A precise image revision | Choose editing controls and review criteria | The change works without unacceptable drift |
| Several visual directions | Organize an exploration process | A direction is selected with reasons |
| A finished video | Choose a production workflow such as Pexo | Visuals, sequence, message, and sound are approved |
How should you interpret launch rankings?
A leaderboard result describes a particular evaluation, not a universal guarantee for your own assets. Before using a ranking to make a production decision, check the evaluated task, date, and comparison conditions. Image generation and image editing are related but distinct tasks; strength on one does not automatically answer every question about the other.
This article is a documentation-based explanation, not a hands-on benchmark. The most useful next step is a small representative trial with a written acceptance checklist. Keep the input, selected service, model identifier, and review notes. That record will make a later comparison meaningful when a preview changes or a new model arrives.








