How to compare text removal tools without fooling yourself

Every remover claims “clean” and “fast.” Most of those claims are untestable marketing. Here is a small protocol you can run in an afternoon — fair to each tool, hard on hype, including ours.

Build a tiny fixture set

• Five to ten real photos you actually care about — not the vendor’s hero shot.
• Mix surfaces: flat colour, brick or fabric, product label, screenshot UI, one hard FX case (glow or thick stroke).
• Keep originals read-only; work on copies.

Score what you can see

For each tool and each photo, look at 100% crops of the erase region and of a neighbour that should not have changed. Pass / fail is enough; skip invented 0–100 scores.

• Ghosting — can you still read the word in the grain?
• Neighbour damage — did a logo, face, or second line soften?
• FX — did stroke / shadow / glow leave with the letters?
• Mask effort — pixel outline required, or a rough box enough?
• Time — wall-clock from upload (or open) to a usable export.
• Trust — retention stated in writing, or silence?

What not to do

• Do not trust a single viral demo with unknown lighting and unknown edits.
• Do not average “quality” across tools that never saw the same file.
• Do not quote images-per-minute you did not measure on your own hardware and network.
• Do not treat “AI” as a score — specialised text erase and generic object remove are different jobs.

Where EraseText should win on that sheet

On lettering with FX, forgiving masks, no ghosting, and a few-second turnaround — including about a second for the erase pass itself. If another tool beats us on your fixtures, keep it. Bring the crop; we would rather lose a fair test than win a vague one.

Run the test

Put the same five files through EraseText and whatever you use today. Keep the winners. Discard the slogans.