Test a result, then reproduce it
The lab explains a specific result from the actual editor. Each published case includes its source, settings, untouched downloads, observed defects and reproduction steps.
Maintained by Arunvpp. The examples are individual tests, not a ranking of model accuracy or a promise about your photos.
Illustrated logo · plain background
Why does the paper stay inside a letter?
Tested . Site-provided illustration; not a photograph or customer result.

Existing SansBG logo sample, described in the site credits as drawn for this site. Source credits. Reused from the existing site sample for this demonstration. No separate license to redistribute the artwork is offered.
Settings and test environment
Chrome 156.0.8078.47 on Linux, desktop viewport 1440 × 1000. Plain background runs in the pixel worker; no AI model or GPU was used.
- Method
- Plain background: Auto, then Remove a color
- Strength / Edge softness
- 25 / 10 for both results
- Color picker for Remove a color
- #fbf9f5, measured from the source corner (251, 249, 245). Set explicitly; the picker initially uses green.
- Export
- None background · Original size · PNG · 1200 × 800 pixels
- Finishing
- Crop to subject off · shadow off · no brush edits
Auto: keep enclosed matching areas

The outer paper clears. Off-white regions enclosed by the letters remain opaque because they are not connected to the image border. This can be useful for light details that belong to a subject.
Manual correction: None. This is the PNG downloaded from the editor.
Download this actual PNG resultRemove a color: clear matching areas throughout

The same enclosed off-white regions become transparent. The colored symbol remains at the checked point. Global color removal can also erase matching colors that were meant to stay in an illustration.
Manual correction: None. This is the PNG downloaded from the editor.
Download this actual PNG resultWhat the saved pixels show
Alpha ranges from 0 (transparent) to 255 (opaque). These are measured points in these specific exports.
| Pixel location | Plain background · Auto | Plain background · Remove a color |
|---|---|---|
| Outer corner (0, 0) | Alpha 0 (transparent) | Alpha 0 (transparent) |
| Inside R (390, 545) | Alpha 255 (opaque) | Alpha 0 (transparent) |
| Inside b (645, 570) | Alpha 255 (opaque) | Alpha 0 (transparent) |
| Inside d (815, 570) | Alpha 255 (opaque) | Alpha 0 (transparent) |
| Yellow symbol (600, 325) | RGBA 241, 194, 77, 255 | RGBA 241, 194, 77, 255 |
Reproduce this comparison
- Open the logo workflow linked below, then choose its Logo sample. This uses the same 1200 × 800 input shown here.
- Choose Plain background → Auto. Set Strength to 25 and Edge softness to 10. Keep Background on None, Size on Original and format on PNG; turn crop and shadow off.
- Download the PNG. Reopen it on a dark background and inspect the enclosed parts of R, b and d.
- Switch to Remove a color. Set its color picker explicitly to #fbf9f5; leave Strength and Edge softness unchanged. Download a second PNG.
- Compare the two saved files at their actual size. A pixel inspector can reproduce the alpha values below; coordinates start at the top-left of the 1200 × 800 file.
Open the logo workflow Download the source sample
Limits and conclusion
- One flat-color illustration tests border connectivity and color matching. It says nothing about portrait, fur, glass or AI model accuracy.
- The sample has soft raster edges and slightly varying backdrop colors. Inspect small lettering at full size: partial edge pixels remain, and neither method reconstructs a clean vector outline.
- A different source, strength, edge setting, browser decoder or editor revision can change individual pixels and output hashes.
- The checkerboard is a page display aid, not part of the PNG. No processing-duration benchmark was taken.
Use Auto when enclosed matching regions should remain. Use Remove a color when every matching region, including letter counters, should become transparent; check that it has not removed wanted light details. For precise brand artwork, prefer an original transparent or vector asset when available.
Check the source and result fingerprints
SHA-256 identifies the exact files used in this case; it does not measure image quality.
- Source
f7007299b54b723040a88b17428a31d0b6fa0a408974f9725c3fb803a8de7320- Plain background · Auto
f1c23851f5cb6b0984d5e94c7f890ad5990aeb1fc1c19b1ffc39fcb9d6bc8284- Plain background · Remove a color
e6e71d17719caf275f94a4193c96124809a4af26f34ce582fce231f2de8c0761
How a case earns a place here
Use a source with documented permission, run the real editor, save the actual output and record the settings. Inspect the saved file at full size, include failures as well as useful results, and label any manual correction. Measure durations only when the method and environment are recorded.
This first comparison uses an existing illustrated site sample. It does not test portraits, fur, reflective products or transparent materials. Further photo cases need suitable source images and permission before publication.