Windows 11 Paint has evolved from a nostalgic doodling canvas into a sophisticated image editor that now thinks in two completely different ways depending on your hardware. The latest updates deliver a clean set of modern editing tools to all users—layers, transparency, project files, and one-click background removal—while simultaneously introducing cloud-powered AI features that burn Microsoft 365 credits. But on Copilot+ PCs equipped with a neural processing unit (NPU), Paint unlocks a local AI co-creator that never calls home and never deducts a credit.
This split-brain architecture is not a bug or a half-baked rollout. It is a deliberate strategy by Microsoft to differentiate the Copilot+ brand while still enhancing the legacy Paint experience for hundreds of millions of existing Windows 11 machines. The result is an app that can feel like two different programs depending on whether you open it on a $300 budget laptop or a $1,200 Snapdragon X Elite ultrabook.
The Universal Foundation: Layers, Transparency, and Canvases That Feel Modern
For years, users begged for a simple layering system in Paint. Microsoft finally listened, adding support for multiple layers and true transparency in late 2023. The implementation is intentionally modest: you can stack, reorder, merge, and hide layers, but you won’t find blend modes, masks, or advanced compositing. Yet even this basic layer panel transforms Paint from a single-layer scratchpad into a mini Photoshop lite for quick memes, mockups, and photo collages.
Alongside layers, Paint adopted a project file format (.proj) that preserves your entire layer stack and editing history. Previously, closing the window meant flattening everything forever. Now you can save your work-in-progress, revisit it days later, and continue tweaking without losing individual elements. The traditional save-as-image options remain, but the project format is a genuine productivity boost for anyone who iterates.
Background removal arrived as another table-stakes feature. Click the dedicated button or hit Ctrl+Shift+X, and Paint uses on-device machine learning—no cloud required—to detect and delete backgrounds in seconds. It works well on portraits and objects with clear edges, though it can struggle with hair, fur, or low-contrast scenes. Still, for a tool that was notorious for its eraser smudge, this is a revelation.
These three features—layers, project files, and background removal—are available to every Windows 11 user running the modern Paint app. They do not require a Microsoft 365 subscription, an NPU, or an internet connection beyond the initial delivery via the Microsoft Store. They represent the “old brain” of Paint: upgraded, but still fully local and free.
The Cloud Brain: AI Image Creation That Costs Microsoft 365 Credits
Layered on top of the base editor is an entirely different AI engine. Microsoft has embedded a generative image creator powered by OpenAI’s DALL-E technology directly into Paint’s toolbar. On any internet-connected Windows 11 PC, clicking the “Cocreator” icon opens a prompt box where you can describe an image and see AI-generated results populate your canvas.
This feature does not run locally. Your prompt, style selections, and iterative refinements are sent to Microsoft’s servers, which return one or more image variations. Crucially, each generation consumes an AI credit from your Microsoft 365 account. The credit system, first introduced alongside Microsoft Designer and Copilot integrations, is now unified across Paint, Photos, Designer, and other Microsoft apps.
Under the current credit structure, Microsoft 365 Personal and Family subscribers receive 15 AI credits per month. Free Microsoft account holders get 15 credits as a one-time welcome before being prompted to subscribe. Additional credits can be purchased through the Microsoft 365 admin center, though consumer pricing for credit packs has not been broadly advertised. Each credit roughly equals one image generation request; higher-resolution outputs or multiple variations may consume more.
This pay-per-generation model has sparked predictable pushback. Enthusiasts who once used free AI image generators like Bing Image Creator (which still exists) now find the experience gated inside their primary editing app. However, Microsoft argues that integrating DALL-E directly into Paint streamlines the workflow: instead of generating an image in a browser, downloading it, and then opening it in Paint, everything happens in one window. You can even use AI-generated elements as layers atop manual sketches, blending generative and traditional editing in a single canvas.
The cloud AI also introduces a quality-of-life feature called “Fill” that uses inpainting to expand or modify areas of an existing image. Select a region, describe what you want, and the cloud model synthesizes new pixels that match the surrounding context. It’s a potent tool, but every fill costs credits—and it requires a steady internet connection.
The Local NPU Brain: Cocreator on Copilot+ PCs
If you buy a Copilot+ PC—currently devices powered by the Snapdragon X Elite, Snapdragon X Plus, or the latest Intel Core Ultra and AMD Ryzen AI 300 series processors—Paint reveals an entirely different AI personality. On these machines, the Cocreator button does not send prompts to the cloud. Instead, it taps into the device’s neural processing unit to run a locally stored AI model that transforms your finger-painted doodles and text descriptions into finished artwork.
This local Cocreator was the headliner feature at the Copilot+ launch in mid-2024. The workflow is designed to feel like a collaboration: you draw a rough sketch—maybe a stick-figure cat or a lumpy car—and type a description like “orange tabby cat sleeping on a rug in sunlight” or “vintage red convertible driving along a coastal highway.” The model blends your sketch’s composition and color palette with the textual concept, generating multiple variations that you can adjust with a creativity slider. Push the slider toward “more creative” and the output drifts further from your original lines; slide it toward “more precise” and it clings more tightly to your brushstrokes.
Because the model runs entirely on the NPU, the generation feels nearly instantaneous—typically one to two seconds for a 1024×1024 output. There is no credit deduction, no internet requirement, and no data leaving your device. For privacy-conscious artists, educators, or business users manipulating sensitive diagrams, this is a critical differentiator.
The technical underpinnings rely on a diffusion model that Microsoft has distilled and quantized to run efficiently on the 40+ TOPS NPUs mandated for Copilot+ certification. Early benchmarks show the model holds up well against cloud-based DALL-E for simple scenes, but it can struggle with complex compositions, photorealistic faces, or text rendering—areas where larger cloud models still have an edge. Microsoft has not disclosed the exact model architecture, but it appears to be a custom version of the Stable Diffusion family, fine-tuned on a dataset that aligns with Microsoft’s content safety filters.
VentureBeat’s hands-on at the initial demo noted that the local Cocreator “feels like magic the first time you see it blend a childlike drawing into a polished illustration,” but also observed that “pro artists will quickly hit the model’s ceiling and may want to export the base for further refinement in Photoshop.” The tool is clearly aimed at casual creators, students, and anyone who wants to visualize an idea without learning professional design software.
Developer and Enterprise Implications
The split architecture carries significant implications for IT administrators and developers. On managed Windows 11 fleets, the cloud AI features can be disabled via Group Policy or Intune. The “Allow Generative AI in Paint” policy setting controls whether the cloud-based Cocreator (formerly called “Paint Cocreator” or “Azure AI integration”) appears. Local NPU features, however, are governed by a separate “Allow AI on Device” policy. This gives admins the ability to permit secure, offline AI generation while blocking cloud data egress.
Microsoft has not yet released an SDK for the local Cocreator to third-party developers, but the underlying Windows Copilot Runtime—which includes the NPU-optimized model—is part of the broader AI infrastructure that will eventually support APIs for any Windows app. When that happens, expect Paint’s local AI to serve as a reference implementation that developers can integrate into their own graphics tools.
What’s Still Missing
Despite all the upgrades, Paint remains an odd beast. There is still no text-on-path tool, no proper gradient editor, no non-destructive adjustment layers, and no support for RAW photo files. The brush engine, while improved, lacks pressure sensitivity that would make it useful with a stylus on tablets like the Surface Pro. The selection tools feel dated next to even basic competitors like Paint.NET or GIMP.
Moreover, the two AI brains do not communicate. You cannot, for example, use the cloud DALL‑E generator to create a starting point and then refine it with the local NPU model—each operates in its own silo. A project file cannot store both cloud-generated assets with their prompt history and local Cocreator iterations in a unified, fully repopulating document. If you generate a cloud image, decorate it with layers and local sketches, and then reopen the project on a different PC, only the local layers survive; the AI-generated raster is baked into a flat image.
Microsoft has hinted that future updates will unify the creation flow and allow a seamless handoff between cloud and local generation, but no timeline has been given. In the meantime, users must choose their AI path based on hardware and connectivity.
How Users Are Responding
Community forums and Reddit threads paint a picture of cautious excitement tempered by confusion over credits. Many casual users do not realize that the “Cocreator” button in Paint on a standard PC is cloud-backed until they exhaust their monthly credit allotment and are prompted to upgrade their Microsoft 365 plan. Microsoft displays a small credit counter in the Cocreator panel, but it is easy to overlook.
Early Copilot+ adopters who reviewed the local Cocreator praised its speed and appreciated not burning credits, but some expressed disappointment that the model cannot generate images beyond 1024×1024 resolution. Others noted that the creativity slider’s “more precise” mode still produces noticeable deviations from their original sketch, making it unsuitable for technical illustrations where exact shapes matter.
The addition of layers and project files has been universally praised, with many longtime Paint users calling it the update they’ve waited decades to see. The free background removal tool has also been a hit, particularly among real estate agents, e‑commerce sellers, and social media managers who need quick product cutouts without launching a heavy editor.
The Competitive Landscape
Paint’s transformation fits into Microsoft’s broader push to make Windows the platform for AI-accelerated creativity. Apple’s macOS offers on-device generative AI through its Neural Engine and apps like Pixelmator Pro and upcoming macOS features, but those tools are not bundled with the operating system. Google’s Chromebooks are receiving AI-powered photo editing tools, but they remain cloud-dependent. Paint’s unique selling point is that it delivers both—a free local AI on premium hardware and a freemium cloud AI on everything else—in an app that has been part of Windows since 1985.
Adobe, Canva, and Figma are not losing sleep over Paint’s AI ambitions, but the strategic threat is real: if a built-in OS app can handle 80% of casual image tasks, why pay for a Creative Cloud subscription or a Canva Pro account? Microsoft is betting that seamless integration—plus the sheer ubiquity of Paint—will siphon off the long tail of users who never wanted complexity in the first place.
Looking Ahead
The current builds are clearly a v1. Microsoft’s Windows roadmap, glimpsed through Insider builds and job listings, suggests active development on several fronts. A vector layer type could finally bring basic shape manipulation without rasterizing. An optional “pro” mode might unlock higher-resolution canvas sizes and pressure sensitivity for stylus users. API hooks could let third-party AI models plug into Paint’s inference pipeline, potentially opening the door to Stable Diffusion XL or Adobe Firefly integrations.
More immediately, Microsoft is expected to refine the credit experience. The company may introduce a “credit-free weekend” or a limited number of free cloud generations per month for all users to lower the barrier to entry. The boundary between what runs locally and what requires the cloud could also blur as NPUs become powerful enough to run larger models—though that transition is likely two to three silicon generations away.
For IT buyers, the Paint story adds another variable to hardware purchasing decisions. If your workforce regularly uses image generation for brainstorming, marketing, or training materials, the local Copilot+ capability could offset subscription costs over time. Conversely, if your organization blocks cloud AI for compliance reasons, enabling the local Cocreator on Copilot+ hardware becomes a rare on-ramp to generative AI within a locked-down environment.
Windows 11 Paint’s split-brain strategy may seem disjointed at first glance, but it reflects a pragmatic reality: the PC ecosystem is too diverse for a one-size-fits-all AI model. By building a hybrid app that adapts to the silicon underneath, Microsoft is hedging its bets and laying the groundwork for an AI-powered Windows future that doesn’t leave legacy hardware behind. Whether users embrace the two-faced canvas or yearn for a simpler, pre-AI Paint will depend on how gracefully Microsoft balances free features, credit pressure, and the genuine magic of generating art from a rough sketch.