Interactive reconstruction laboratory
See what AI imagines beyond the anchor.
Upload an image, preserve one region, and let an image model imagine everything else. Then compare its reconstruction with the original.
Start an experiment
Private by design Measured in-browser
01SelectPreserve a visual anchor
02ReconstructGenerate the missing context
03CompareInspect objective metrics
Example reconstruction
See what AI can infer from a glimpse.
One preserved region gives the model visual context to create a plausible continuation of the complete scene.
A plausible reconstruction, not retrieval. AnchorLens generates a best-effort completion from visible visual cues; it does not recover a stored original image.
Try your ownExperiment setup
Choose your visual anchor
Drop an image here
PNG, JPG, or WebP · maximum 8 MB
1
Draw the anchor
Drag over the detail AI should keep
2
Anchor + noise
Everything outside the anchor is hidden
Guest keys are sent to this server for the current request and are not saved by AnchorLens.
Member access
Your account includes 3 reconstructions per UTC day.
Draw an anchor to continue.
Preparing reconstruction
Building the masked image for the model...
Comparison
What did the model recover?
Full image
Original and reconstruction compared across every pixelOverall similarity
Regenerated region
Only pixels outside the preserved anchor are comparedMissing-area similarity
Automated interpretation
—
A conclusion will appear after the comparison is calculated.