WebcamTest.com
← Webcam Test home

Client-Side AI Segmentation

Background Blur & Edge Stability Test

Video apps blur or replace your background using person segmentation — and it often glitches around hair, or lags on older hardware. This runs the same kind of model directly in your browser (no upload, nothing sent anywhere) so you can see exactly how clean your setup's edges are before you're on a call.

Camera off

Controls

Effect Settings

18px

Readout

Segmentation Quality

Click "Start Test" to begin. Scoring needs a few seconds of frames to stabilize.

Edge Stability Score

—/100
Frame-to-Frame Stability

How much your mask's edge flickers when you hold still. Higher is cleaner — a low score usually means visible "shimmer" around your outline on a real call.

Artifact Rating

—
  • Speckle Count (per frame, avg):—
  • Foreground Coverage:—
  • Segmentation FPS:—
  • Compute Delegate:—

Artifact rating is a heuristic estimate of isolated misclassified pixels (halo/speckle noise), not a certified image-quality metric.

Why It Happens

Reading Your Results

Low stability + high speckleImprove lighting contrast between you and the background — segmentation models struggle most in flat, low-contrast lighting.
Halos around hair specificallyCommon and hard to fully avoid — try pulling hair back or adding a light behind you to add edge contrast.
Score drops when you moveExpected — motion blur makes edges harder to classify frame to frame. A stress test on purpose.

Benchmark Reference

Stability & Artifact Bands

MetricGoodFairPoor
Edge stability score≥ 7040 – 69< 40
Speckle / artifact rating< 2 avg2 – 7 avg≥ 8 avg

These are this tool's own real scoring thresholds, computed frame by frame while you're live.

Step-by-Step

How to Run the Test

  1. Click "Start" and allow camera access when prompted, then wait a moment for the AI segmentation model to load.
  2. Hold still for a few seconds to let the stability score settle to a representative baseline.
  3. Move naturally — turn your head, gesture — to see how much the score drops under realistic motion.
  4. Adjust the blur amount slider and watch how stability and edge artifacts change together.

Common Issues

Troubleshooting & Fixes

Camera access blockedClick the camera icon in your browser's address bar to confirm this site is allowed, or check your OS's camera privacy settings.
"Model failed to load"Check your internet connection — the segmentation model loads from a CDN on first use and needs that connection to succeed.
Score stuck low even sitting stillCheck the delegate readout — CPU fallback (WASM) is slower and can affect scoring smoothness on machines without GPU acceleration available to the browser.
Background and clothing similar in colorLow contrast between you and your background is the single biggest driver of poor segmentation — a different shirt or a brighter background light both help measurably.

Share

Share Your Result

Related Tools

Take This Further

AI segmentation is only one route to a clean background. If you own a green screen, test a physical green screen instead — a well-lit screen still outperforms software segmentation at the edges.

Common Questions

Frequently Asked Questions

Why does my virtual background look bad around my hair?
Hair has thousands of fine edges that segmentation models struggle to classify frame to frame. More contrast between you and your background helps most; tying hair back or adding a light behind you also improves the edge.
How do I test if background blur is working properly?
Start the test and watch the edge stability score. It measures how consistently your silhouette holds up between frames, which is what actually determines whether blur looks clean or flickers.
Does background blur need a green screen?
No. AI segmentation works without one, which is the point. A physical green screen still produces a cleaner edge if you have the space and can light it evenly.
Why does background blur get worse when I move?
Motion blurs the edges the model relies on, so classification becomes harder frame to frame. Some degradation during movement is normal on every platform, not a fault with your camera.
Does this send my video anywhere for the AI to analyze?
No. The segmentation model runs entirely on your device using WebGL or WASM — nothing is ever uploaded.
Is GPU or CPU delegate better for this test?
GPU is faster and generally scores more smoothly since it can process more frames per second, but CPU (WASM) fallback still works correctly — just check which one your results actually used before comparing scores across different devices or sessions.