is there any mode to also remove text from the image?

[danyhm]

Hello,
some service online offer text + logo removal as well.
like unwatermark.ai
watermarkremover.io
and a few more.
i want to process 10M+ images and it's not affordable to use their services.
any recommendation ?

[D-Ogi]

Hi @danyhm,

Great question! Yes, the tool can remove text - you just need to use a custom detection prompt:

python remwm.py input.png output/ --detection-prompt="text"

Or for both text and logos:

python remwm.py input.png output/ --detection-prompt="text logo watermark"

For your scale of 10M+ images, here are some recommendations:

1. Use CLI mode - it's more efficient for batch processing than the GUI
2. Process in batches - point the tool at folders:

   python remwm.py ./input_folder ./output_folder --overwrite

3. GPU acceleration - if you have CUDA-capable GPU, processing will be significantly faster
4. Consider detection-skip for videos - if any of your content is video, --detection-skip=3 speeds things up

The detection quality depends on Florence-2's ability to recognize the specific text/logo patterns. For consistent watermarks across millions of images, results should be reliable. For varied watermarks, you might need to experiment with different detection prompts.

Would you be willing to share a sample image? I could suggest the optimal settings for your specific use case.

[danyhm]

Hi @D-Ogi
Thank you very much for your extended reply!
Here is a sample.
sample.zip
This is the worst-case scenario for me. (logo + text to remote + text to keep)

What I want to do is (in order of preference):

1. Remove the logo watermark in the center of the image.
2. Remove the URL text and other text from the blue region (not all the images have this URL in them, but the position of it is almost fixed in all of them)
3. Keep the ruler scale text (e.g., 0cm, 1cm)