Microsoft has quietly begun rolling out KB5064647, an automatic Windows Update that upgrades the Image Transform AI component on Copilot+ PCs to version 1.2507.793.0. The update sharpens the ability of these AI-accelerated machines to erase foreground objects from photos and fill the gaps with generated backgrounds—a feature that until recently required desktop image editors and cloud processing. With this release, the company replaces the earlier KB5063136 and brings what it describes as “improvements” to the component, though it stops short of detailing the specific refinements.
For the millions of Windows 11 24H2 users running Snapdragon X Elite or other supported NPU-powered notebooks, the change will arrive through Windows Update without any manual intervention, provided they have kept their cumulative updates current. The move is the latest in a series of under-the-hood AI component refreshes that arrived alongside the 2024 update and positions on-device AI as a default creative tool rather than an optional extra.
What the Image Transform AI component actually does
The Image Transform AI component is a system-level imaging service designed specifically for Copilot+ PCs—the Qualcomm Snapdragon and eventually Intel Lunar Lake and AMD Strix Point devices with a dedicated neural processing unit. When a user selects a foreground subject in a photo, the component runs an on-device model to separate the subject from the background, remove it, and then synthesize realistic pixels to fill the void. The result looks like the object was never there, with shadows, textures, and lighting matched to the scene.
The work happens entirely on the device. No images are uploaded to the cloud, which addresses two perennial complaints about AI editing: latency and privacy. Microsoft had previously demonstrated this capability in the Photos app for Copilot+ PCs, but the Image Transform AI component serves as a shared engine that other applications can tap. Updates to the component therefore improve background filling and object erasure across any Windows app that uses the API, not just Photos.
What’s new in version 1.2507.793.0
Microsoft’s support note is deliberately sparse. It says only that the update “includes improvements to the Image Transform AI component for Windows 11, version 24H2.” The previous version, replaced by KB5064647, was KB5063136. The lack of a detailed changelog is not unusual for Windows AI component updates, which often receive iterative model refinements. Based on earlier whispers in developer channels, these improvements likely center on edge detection and hole-filling aesthetics. Early builds sometimes left visible seams or struggled with complex backgrounds like foliage or patterned walls. The new version may be more robust with challenging foreground shapes, translucent objects, or areas where the subject casts a strong shadow.
Community members on Windows forums have already noticed the update appearing in their update history. One early thread compiled the key details from Microsoft’s support page, prompting speculation about whether the improvements extend to video scenarios or remain strictly for still images. Microsoft has not confirmed any video capabilities, but the API and the underlying NPU pipeline are architected in a way that could eventually support frame-by-frame removal in near real time.
How to get KB5064647
KB5064647 will download and install automatically through Windows Update as long as the device meets the prerequisites:
- The PC must be a Copilot+ PC with a compatible NPU.
- Windows 11 version 24H2 must be installed.
- The latest cumulative update for 24H2 must already be present on the system.
There is no need to hunt for a standalone download or trigger an early install from the Microsoft Update Catalog. For managed enterprise environments, the update will flow through Windows Update for Business and WSUS, respecting existing deferral policies. IT admins should note that while the component itself is not security-related, it will appear in the update history alongside optional quality updates.
To verify whether KB5064647 is installed, open Settings → Windows Update → Update history. Under the “Other updates” section, you’ll see an entry labeled:
2025-07 Image Transform version 1.2507.793.0 (KB5064647)
The listing is processor-specific—Snapdragon devices will show a Qualcomm-tailored update description, while future x86 Copilot+ devices may display a slightly different string upon their release.
Which editions get the update
Microsoft has published the list of applicable editions, and it includes virtually every 24H2 variant:
- Windows 11 Home
- Windows 11 Pro
- Windows 11 Enterprise
- Windows 11 Education
- Windows 11 Enterprise Multi-Session
- Windows 11 SE
- Windows 11 IoT Enterprise
The broad distribution signals that Microsoft sees the AI imaging stack as a core feature of the Copilot+ platform, not a niche creative tool. Whether you own a $999 consumer laptop or use a hardened IoT thin client, the component will be on your disk and capable of being called by any qualifying application.
What KB5064647 replaces
The update directly replaces KB5063136, the February 2025 release that first stabilized the Image Transform AI component for public use. That earlier version resolved some initial compatibility issues with early Copilot+ SKUs and was itself a prerequisite for several third-party image apps that began shipping with Windows Studio Effects and other AI capabilities. When KB5064647 lands, the previous KB number is removed from the update history, and the component is upgraded in place.
The engine beneath the magic: NPU and DirectML
To understand why this update matters, it helps to peek under the hood. Copilot+ PCs ship with an NPU capable of at least 40 trillion operations per second (TOPS). The Image Transform AI component leverages that silicon through DirectML, Microsoft’s low-level API for machine learning inference on Windows. When a foreground selection is made, the task is routed to the NPU, which runs a segmentation model (likely a U-Net variant) to produce a precise alpha mask. Then an inpainting model—probably a lightweight convolutional network—generates the background pixels. The whole pipeline can complete in under two seconds on a 12-megapixel image, a dramatic improvement over CPU-bound solutions that can take ten times as long.
Version 1.2507.793.0 may ship with an updated ONNX model package that further optimizes inference. Such tweaks can yield better memory usage, faster tile processing, or higher-resolution masks without altering the public API. Developers who have already adopted the Image Transform capability should see the benefits transparently, with no code changes required.
The broader AI push in Windows 11
KB5064647 is not an isolated patch. It follows the Architecture component updates for Studio Effects, Voice Clarity, and Windows Copilot that have been trickling out since the 24H2 launch. Each of these updates improves on-device AI experiences without requiring a full OS upgrade. Microsoft is essentially building an AI operating system in layers: the foundational NPU driver stack, then the componentized AI services (like Image Transform), and finally the user-facing features in Photos, Paint, and Camera.
This modular approach allows Microsoft to ship improvements on a monthly cadence rather than waiting for a semi-annual feature release. It also means that when Windows 11 25H2 eventually arrives, the AI stack will already be mature, and the new version can focus on UI polish and new scenarios rather than playing catch-up with model performance.
Community reception and real-world impact
Windows enthusiasts on forums have largely welcomed the update, though some ask when a comparable set of AI tools will arrive for non-Copilot+ PCs. Microsoft remains quiet on that front. The official stance is that the NPU is required for these experiences to meet quality and reliability bars. Emulating the models on a GPU or CPU would drain battery and bring the interactive fluidity down to unacceptable levels. For now, the 40+ TOPS NPU is the price of admission for on-device background erasure.
One practical discussion point that emerged from forum threads is the prerequisite chain. Users who defer cumulative updates for testing must be aware that they won’t receive KB5064647 until they install the latest 24H2 cumulative update. Some enterprise admins inadvertently blocked the December 2024 cumulative update and saw Image Transform stuck on an older version. The fix is simple: allow the latest cumulative update, after which the AI component flows down automatically.
Comparing with cloud alternatives
Before Copilot+ PCs, a similar task in Microsoft ecosystem would have required uploading the photo to OneDrive and using the cloud-backed Generative Erase in the Photos app. That process took several seconds of latency and generated privacy concerns. Third-party tools like Adobe Photoshop (via the Firefly-powered Remove Tool) and Canva’s Magic Eraser also rely on server-side processing in many configurations. The Windows 11 approach keeps everything local, which is not only faster but also makes the feature available offline—a meaningful advantage for photographers working in the field.
KB5064647 thus reinforces Microsoft’s “AI for every PC” narrative for Copilot+ devices. Rather than being an also-ran, the built-in capability is now competitive with the best cloud services for single-object removal, at least on simple backgrounds.
What to expect next
The update history list for Copilot+ PCs is becoming a fascinating log of AI evolution. In the coming months, we may see similar refreshes to the Photos app’s Restyle feature, which also relies on on-device models, and to the Video Trim tool that hints at future video inpainting. Microsoft hasn’t announced a roadmap, but the existence of a dedicated Image Transform component strongly suggests that the company wants to treat AI imaging as a platform capability rather than a point feature in one app.
For now, Windows 11 24H2 users on Copilot+ PCs should keep their updates turned on and check for KB5064647. It’s a small download that requires no configuration, yet it brings a tangible polish to one of the fastest-growing use cases for AI accelerators. Object removal may sound like a niche feature, but in a world of polished social media feeds and instant visual communication, a tool that makes bad photos good in seconds has genuine mass appeal. With this update, Microsoft makes that tool a little bit smarter.