How to Remove a Watermark from a Photo Using Neural Networks

Watermarks on photos are a common problem. Sometimes you need to remove your own outdated watermark, restore an old photo, or clean up a screenshot. Neural networks have learned to do this in seconds. But before moving on to the tools, let's talk about something important—ethics and legality.

Ethics and Legal Issues

When Watermark Removal is Acceptable

• Your own photos — you applied the watermark and want to remove or replace it.
• Photos with an expired license — you purchased a license, but the original file without the watermark is lost.
• Personal photos — someone added their watermark to your photo without permission.
• Testing and learning — you are studying AI capabilities using your own images.

When Watermark Removal is Unacceptable

• Commercial stock photos — the watermark protects the photographer's copyright.
• Others' work — removing a watermark from someone else's photos violates copyright.
• Commercial use — using "cleaned" photos belonging to others for business is illegal.

Removing watermarks from others' content can lead to legal consequences under copyright law. Use these tools responsibly.

Cleanup.pictures — The Simplest Option

Cleanup.pictures is a free online tool that removes any unwanted objects from photos, including watermarks.

How to Use

1. Go to cleanup.pictures.
2. Upload an image (drag and drop or click the upload area).
3. Select the "Brush" tool.
4. Paint over the area with the watermark.
5. The neural network will automatically fill the area with appropriate content.
6. Download the result.

Tips for Better Results

• Paint over the watermark with a slight margin around the edges.
• For complex backgrounds, process in parts, not all at once.
• If the result isn't perfect, undo and try painting slightly differently.
• Quality depends on background complexity: uniform backgrounds are processed better.

Free Version Limitations

• Maximum output resolution is 720px.
• Full resolution requires a ClipDrop Pro subscription.
• Number of processing attempts is unlimited.

Inpaint (theinpaint.com)

Inpaint is a specialized tool for removing objects from photos. It works on the principle of "smart fill."

How to Use

1. Open theinpaint.com.
2. Upload a photo.
3. Highlight the area with the watermark using the marker.
4. Click Erase.
5. The algorithm will fill the area based on surrounding pixels.
6. Repeat for any remaining traces if necessary.

Features

• Works without registration.
• Processes images locally (does not send to a server).
• Supports batch processing in the desktop version.
• Handles semi-transparent watermarks better.

Adobe Photoshop — Generative Fill

Adobe has added the Generative Fill feature to Photoshop, based on the Adobe Firefly model. It's one of the most powerful tools for object removal.

How to Use

1. Open the image in Photoshop (version 25.0 or higher).
2. Select the area with the watermark using the Lasso Tool or Rectangular Marquee.
3. On the panel that appears, click Generative Fill.
4. Leave the text field empty (Photoshop will automatically fill the area).
5. Choose the best of the three suggested options.
6. Generate more variations if necessary.

Advantages of Generative Fill

• Highest quality fill.
• Works with backgrounds of any complexity.
• Generates on a separate layer (does not destroy the original).
• Multiple options to choose from.
• Supports text prompts to guide generation.

Disadvantages

• Requires an Adobe Creative Cloud subscription.
• Needs a powerful computer or stable internet.
• Processing is done in the cloud (privacy consideration).

DALL-E Inpainting via ChatGPT

OpenAI allows the use of DALL-E for inpainting—replacing a selected area of an image.

How to Use

1. Open ChatGPT (the version with DALL-E).
2. Upload an image to the chat.
3. Ask it to remove the watermark:

This image has a watermark in the bottom right corner.
Remove it and fill the area, preserving the natural background.

4. DALL-E will generate a new version of the image.
5. Refine your request if necessary.

Limitations

• The result may differ from the original (DALL-E "redraws" the entire image).
• Not suitable for photos where detail accuracy is crucial.
• Works better with simple backgrounds and compositions.
• Quality depends on the prompt.

Lama Cleaner — Free Local Processing

Lama Cleaner is an open-source tool for removing objects from images. It works locally on your computer, addressing privacy concerns.

Installation

pip install lama-cleaner
lama-cleaner --model=lama --device=cpu --port=8080

After launching, open http://localhost:8080 in your browser.

How to Use

1. Install Lama Cleaner via pip (Python 3.8+).
2. Launch the local web server.
3. Open the interface in your browser.
4. Upload an image.
5. Paint over the area with the watermark using the brush.
6. The neural network will process the image locally.

Available Models

Advantages of Lama Cleaner

• Completely free.
• Works offline—your data never leaves your computer.
• Supports multiple AI models.
• No limits on quantity or resolution.
• Open source.

Comparison of All Tools

Legal Alternatives to Stock Photos

Instead of removing watermarks from paid stock photos, consider free sources of quality images:

Photo Stocks with Free Licenses

• Unsplash — over 3 million high-quality photos, free for commercial use.
• Pexels — free photos and videos from a community of photographers.
• Pixabay — over 2 million free images, illustrations, and videos.
• Burst by Shopify — free photos for business and e-commerce.
• StockSnap.io — hundreds of new photos every week with no restrictions.

Image Generation with Neural Networks

Instead of searching for a suitable stock photo, create a unique image using AI:

• Midjourney — high-quality, photorealistic images.
• DALL-E 3 — generation via ChatGPT, works well with text descriptions.
• Stable Diffusion — free, open-source model, can be run locally.
• Kandinsky (from Sber) — free generator with good understanding of Russian language.
• Shedevrum (from Yandex) — free app for generating pictures.

Tips for Better Watermark Removal

Image Preparation

• Work with the maximum available resolution—more pixels mean more accurate restoration.
• If the watermark is semi-transparent, try increasing the contrast first.
• Always keep a copy of the original before processing.

Removal Technique

• Process large watermarks in parts.
• For text watermarks, paint over each letter separately.
• After the main removal, go over any remaining artifacts again.
• Compare the result with the original at 100% zoom.

Post-Processing

• Check the removal area for blurring or artifacts.
• Use sharpening tools if necessary.
• Ensure background textures match the surroundings.

Conclusion

Neural networks have made watermark removal technically simple—it takes literally seconds. For quick results, use Cleanup.pictures; for maximum quality, use Photoshop Generative Fill; for private processing, use Lama Cleaner.

But remember the most important thing: technical capabilities do not override ethics. Use these tools only for your own photos and legally obtained images. For commercial projects, free photo stocks and AI-generated images are the best and completely legal alternative to removing others' watermarks.