What you need
Choose one authorised JPEG, PNG, or WebP still image up to 20 MB and 12 million decoded pixels. Animated frames are not processed.
Image Tools / Local studio
Locate the opacity-weighted center of decoded alpha and report pixel and normalized coordinates without considering RGB brightness.
Step 1 / Source
Or browse for a JPEG, PNG, or WebP image that you own or are authorised to edit.
One file · up to 20 MB · maximum 50 megapixelsStep 2 / Output
This analysis uses one documented method so repeated runs on the same decoded pixels remain comparable.
The selected image is decoded and re-encoded inside this browser tab.
After download, the preview URL, decoded bitmap, canvas, and selected source reference are cleared.
Nirmion cannot remove the finished file from your device or browser download history.
Workflow guide
Choose one authorised JPEG, PNG, or WebP still image up to 20 MB and 12 million decoded pixels. Animated frames are not processed.
Compare transparent icon exports for shifted opacity mass while reviewing optical alignment and visible subject shape separately.
Review and download a separate alpha-centroid JSON report; the original file on your device remains unchanged.
Decision guide
Each nonzero-alpha pixel coordinate is weighted by alpha divided by 255; weighted sums yield canvas and zero-to-one centroid coordinates.
An alpha centroid is not a luminance centroid, subject center, optical center, saliency estimate, bounding box, or alignment recommendation. Browser-local processing avoids a Nirmion file upload, but it cannot remove copies stored elsewhere.
MDN Canvas pixel manipulationGood to know
Questions answered
No. Supported JPEG, PNG, and WebP decoding, pixel inspection, transformation, and export run in your current browser tab. Nirmion does not send the selected image to its API for this workflow. The result is a separate derivative or JSON report, while the original file on your device remains unchanged and should be retained for comparison.
Each nonzero-alpha pixel coordinate is weighted by alpha divided by 255; weighted sums yield canvas and zero-to-one centroid coordinates. Every decoded pixel contributes its alpha weight; fully transparent pixels add zero weight and RGB values are ignored. The method does not infer subjects, restore missing detail, calibrate a display, or develop raw camera data. Compare edges, gradients, text, texture, and transparent areas with the original before relying on the result.
An alpha centroid is not a luminance centroid, subject center, optical center, saliency estimate, bounding box, or alignment recommendation. The workspace accepts one supported still image up to 20 MB and caps this pixel-intensive operation at 12 million decoded pixels for predictable browser memory use. Results depend on browser decoding and encoding support. Animation, raw development, ICC calibration, layer editing, and batch processing are outside this tool's verified scope.