ClearPix is designed to remove watermarks while keeping image and video processing in the browser tab rather than sending media to a processing server. Its developer describes a static-site architecture that downloads AI models to the device, runs them locally, and re-encodes the result in the browser. That can reduce exposure of the media itself, but it is not the same as having no network traffic—and the privacy and performance details below are the developer’s account, not an independent audit.
What “without uploading” means in this design
In Umasou’s account, ClearPix has no processing backend. The browser downloads the application’s runtime and model weights, then decodes, processes, and re-encodes the user’s media in the tab. The intended data flow is therefore different from a server-processed editor: the model comes to the device, instead of the media being sent to a server for inference.
As an Amazon Associate I earn from qualifying purchases.
That does not mean the browser makes no network requests. The developer says aggregate analytics events are collected, while claiming those events cannot contain images, filenames, or file contents. The article also says media-bearing canvas, image, and video elements are masked in Microsoft Clarity session replays. These are the author’s descriptions of the system and its controls, not findings from an independent traffic inspection or privacy audit. Umasou’s 2026 account of ClearPix is the primary source for the implementation claims in this article.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallAs Umasou puts it, “The only bytes that travel are the models coming down to you, and a small set of aggregate analytics events that contractually cannot contain your media.” That is the author’s description; readers should not treat it as a third-party verification of every request or script involved.
#1 Best Overall
- Precise Background Removal
- Object Removal
- Automatic Background Removal
- Fine-Tune Editing Controls
- High-Quality Results
How image processing runs in the browser
Models and runtime arrive at the device
The described editor uses ONNX Runtime Web and downloads model weights from storage or CDN sources. Those files are part of the product’s delivery pipeline: they must be fetched before local inference can run, and they may be cached in IndexedDB for later use. The article’s runtime discussion identifies a roughly 28 MB jsep.wasm dependency, a project-specific figure rather than a general size for browser AI applications.
WebGPU is preferred; WASM is the fallback
ClearPix is described as preferring ONNX Runtime Web’s WebGPU execution provider for inference. If WebGPU is unavailable or proves unreliable, the application can use the WebAssembly (WASM) path instead. This gives the editor an alternate execution route, but it does not guarantee identical speed or reliability across browsers and devices.
The browser decodes, edits, and re-encodes
Once the model is available, the tab handles the media workflow: decoding the input, applying the inpainting operation to a selected region, and encoding the output. That keeps the core media processing on the device in the described architecture. It also means the browser, device memory, graphics support, and local storage are part of the application’s operating environment—not just the interface.
Why local processing still has engineering costs
First use includes downloads and setup
A local editor avoids sending media to a processing server according to the developer’s account, but its models and runtime still have to reach the browser. The first session can take longer while model files download and WebGPU shaders compile. Subsequent use may benefit from cached data, subject to the browser retaining that data and the application validating it.
Rank #2
- Manage and edit your RAW photos with this powerful and ultra-fast photo editor
- With no importing required, conveniently store and organize all your photos in one place
- Load, select, edit and output your photos quickly, with powerful batch processing with controls and presets that allow you to edit one or thousands of photos at once
- Watermark, correct, remove blemishes and more with time-saving integrated tools Image presets automate tedious and repetitive processing work
- Correct distortions introduced by different lenses with built-in support, or share and import new lens corrections with the AfterShot Pro community on Facebook
Large models stress memory and storage
The engineering article describes a 198 MB model whose single structured-clone write reportedly caused renderer out-of-memory trouble. The team says it changed writes above 64 MB to use 32 MB chunks. Those figures describe ClearPix’s implementation, not a general browser limit. They illustrate why moving inference onto a user’s device requires attention to both memory use and the way large files are copied or stored.
Integrity checks protect cached files
According to the author, the application pins SHA-256 hashes for model files and checks cached data to detect corruption. Its delivery logic also chunks large model writes, retries mirrors after stalled downloads, and uses fallback model specifications. These measures address specific distribution and cache failure modes; they do not eliminate the possibility of a failed download, an incompatible device, or a browser-specific problem.
What the reported performance figures do—and do not—show
The ClearPix article reports that its team measured about 2,100–3,500 ms per inpainting region before enabling WebGPU and roughly 400 ms per region afterward. It also reports a delivery comparison of 75–110 KB/s for direct r2.dev URLs versus 2.3 MB/s through a same-zone custom domain. The author says hashing a 67 MB cached model took about 100 ms. These are measurements and engineering observations attributed to the ClearPix team in Umasou’s 2026 article, not independent tests or browser-wide benchmarks. Results on another device, connection, model, or image may differ.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteThe article also says the release archive shrank from 27.4 MB to 10 MB after model and benchmark assets were excluded. That concerns the project’s release packaging; it should not be read as the size of the complete first-run download, since models and runtime assets are separate parts of the described delivery process.
Rank #3
- Non-destructive photo editing: Manage and edit your RAW photos with this powerful and ultra-fast photo editor
- Simple photo management: With no importing required, conveniently view, store and organize all your photos in one place
- Powerful batch processing: Load, select, edit and output your photos quickly with controls and presets that allow you to edit one or thousands of photos at once
- Time-saving tools: Watermark, correct, remove blemishes and more with integrated tools; image presets automate tedious and repetitive processing work
- Get complete control with a personalized workflow: Customizable workspace, easy integration with photo editors, built-in support, and shareable web galleries
What to check when evaluating a “no upload” claim
A statement that an editor works locally is useful, but readers evaluating any such claim should distinguish the media-processing path from the rest of the application’s network activity. Relevant questions include:
- Where do the media bytes go? Is processing performed in the tab, or is the file sent to a server for inference?
- What do telemetry and third-party scripts receive? A local processing claim does not, by itself, establish what analytics or session-replay tools can access.
- What must be downloaded first? Model weights and runtime assets can be substantial, especially on a first visit or after cached data is removed.
- How does the tool behave on less capable devices? Browser graphics support, available memory, and mobile constraints can affect whether processing succeeds.
- What happens when the preferred path fails? A fallback such as WASM can improve resilience, but readers should not assume it works identically or reliably on every configuration.
- How are models and cached files verified? Integrity checks can help detect damaged or unexpected files, while retry and fallback behavior can help with delivery problems.
- Can the privacy claim be independently inspected? An architecture description is informative, but it is not equivalent to an independent review of network traffic and application behavior.
The trade-off: less media transfer, more device dependence
Client-side processing changes who bears the operational burden. The user’s device must download and store models, support the chosen runtime, and provide enough memory and processing capacity. The ClearPix article says WebGPU can fail or stall, missing cross-origin isolation can limit WASM threading, and mobile memory is constrained. It describes a watchdog and WASM fallback as responses to some failures, while acknowledging that hardware is less predictable than a data-center GPU.
The source does not establish that every browser or device works reliably, nor does it provide an independently tested comparison with server-processed editors. It also does not support a market-wide claim that online watermark removers generally upload user media. ClearPix’s design is a concrete implementation account, not proof about other tools.
Recommended Free Tools
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.




