I kept finding nice photos ruined by text stamped right over them. Cropping doesn't help when the text sits in the middle of the picture. So I built VanishText, a Windows app that removes text from images you own or have the rights to edit.
The pipeline
It works in three steps.
- Text detection. The app finds the text in the image, including rotated lines, not only straight horizontal ones.
- Masking. It builds a mask that follows the lettering closely, so only the text area gets replaced and the rest of the image stays untouched.
- Inpainting. An AI inpainting model fills the masked area with pixels that match the surrounding background.
Why local-first
I didn't want anyone's photos going to a server. Everything runs on your PC. There's no account and no telemetry. After a one-time model download of about 0.7 GB, it works offline.
It also runs on ordinary hardware. No graphics card is needed. Batch mode lets you point it at a folder and leave it to work through the images.
Honest limits
- It runs on the CPU, so expect a minute or two per image.
- Very faint text on busy patterns can need a second pass.
- Transparency isn't kept.
- Windows only.
- The installer isn't code signed yet, so SmartScreen may warn you on first launch.
Use it on your own images
It's meant for images you own or have the rights to edit. Please don't use it on other people's photos or copyrighted material without permission.
Price
One-time €19.99, no subscription: https://ko-fi.com/s/0354d9a8db
A question for you
If you've built local-first desktop tools, how did you handle the first-run model download? And what kinds of images would you try this on?
Top comments (0)