How fast is Peace of Mind's image classification?

The InceptionV3 neural network classifies each image in 50-200 milliseconds on typical hardware. Safe images are unblurred almost instantly after the page loads. Because every image starts blurred, there is no visible delay — the blur is simply removed once the image is confirmed safe. The entire process is imperceptible during normal browsing.

AI Image Scanning — How Peace of Mind Blocks Explicit Images (2026)

Pre-Blur: Protection Before Classification

Most image scanning tools follow a see-then-decide approach: the image loads, the AI analyzes it, and then the tool decides what to do. This means the image is visible on your screen — even if only briefly — before any action is taken. For someone in recovery, that brief exposure can be enough to trigger a relapse.

Peace of Mind inverts this process entirely. A 40px Gaussian blur is applied to every image via CSS injected at document_start — the earliest possible injection point in the browser lifecycle. This happens before the page renders, before any JavaScript executes, before the browser has drawn a single pixel of content. Every image starts blurred. You never see an unclassified image.

Once the InceptionV3 model finishes classification, safe images have their blur removed. The transition is smooth and nearly instantaneous on modern hardware. Suggestive images remain blurred. Explicitly harmful images trigger a full page redirect to the block page. At no point during this process is an explicit image visible on your screen.

The InceptionV3 Classification Model

Peace of Mind uses NSFWJS, a neural network built on Google's InceptionV3 architecture. The model classifies every image into five categories:

Porn — Explicitly harmful content. Triggers a block or blur depending on confidence score.

Hentai — Animated or drawn explicit content. Same tiered response as Porn, with an additional Neutral floor check to reduce false positives on safe illustrations.

Sexy — Suggestive but not explicit content. Triggers a blur at moderate confidence. Never triggers a full block — the goal is proportional response, not overreach.

Drawing — Illustrations, cartoons, and artwork that are not explicit. Cleared automatically.

Neutral — Safe content. Cleared automatically and blur removed.

The model uses float16 quantization for a balance of accuracy and performance. The model weighs approximately 43MB and accepts 299x299 pixel input images. Classification takes 50-200 milliseconds per image on typical hardware.

Two-Tier Threshold System

Peace of Mind does not use a single pass/fail threshold. It uses a two-tier system calibrated through extensive real-world testing to minimize false positives while catching harmful content with high accuracy:

Block tier: When the model is highly confident an image is explicit (very high confidence scores), the entire page is redirected to the block page. This is the nuclear option — reserved for content where there is essentially no doubt.

Blur tier: At moderate confidence levels, images are blurred but the page remains usable. This catches suggestive content and borderline images without locking you out of otherwise safe pages. The Sexy category only triggers blur, never block — a proportional response to content that is concerning but not explicit.

The Hentai category includes an additional Neutral floor check: if the model simultaneously reports high Neutral confidence, the Hentai classification is treated as a false positive. This prevents safe illustrations from being incorrectly flagged.

100% On-Device, 100% Private

The InceptionV3 model runs entirely inside your browser. No images are uploaded to any external server. No thumbnails, no hashes, no metadata — nothing leaves your device. The model is loaded once when the extension starts and runs locally for every image on every page.

This is a fundamental architectural decision, not a cost optimization. Privacy in recovery is non-negotiable. Other tools like Covenant Eyes capture screenshots of your browsing and send them to a third party for analysis. Peace of Mind was designed from the beginning to ensure that no one — not us, not a partner, not a server — ever sees what you browse. The AI runs on your machine and the results stay on your machine.

Frequently Asked Questions

How does Peace of Mind scan images for explicit content?

Peace of Mind uses a two-stage approach. First, every image on every page is pre-blurred with a 40px Gaussian blur via CSS injected at document_start. Then, an on-device InceptionV3 neural network classifies each image into five categories. Safe images are unblurred in 50-200ms. Suggestive images stay blurred. Explicit images trigger a full page block.

Does Peace of Mind upload my images to a server?

No. The InceptionV3 neural network runs entirely on your device. No images are ever uploaded to any external server. The model weighs approximately 43MB and is stored locally. All classification happens in your browser.

What is pre-blur and why does it matter?

Pre-blur means every image is blurred before you can see it. A 40px Gaussian blur is applied at document_start, before the page renders. You never see an explicit image even for a split second while the AI is classifying it. Other tools see the image first and report it after the fact. Peace of Mind ensures you never see it in the first place.

How fast is the image classification?

The InceptionV3 model classifies each image in 50-200 milliseconds on typical hardware. Because every image starts blurred, there is no perceptible delay — the blur is simply removed once the image is confirmed safe.