Tutorial
JPEG compression makes watermark removal tricky. Here’s a step-by-step guide using AI tools that actually work on compressed images.
JPEG uses a compression algorithm based on 8x8 pixel blocks. The image is divided into a grid of these blocks, and within each block, the codec discards high-frequency detail that humans are unlikely to notice at a glance.
The problem for watermark removal: watermark text and logos have sharp, high-contrast edges. When JPEG compresses an image with a watermark already applied, those sharp edges create distinctive compression artifacts — called ringing and blocking — that extend into the surrounding pixels. The watermark and its compression artifacts become entangled with the background.
This means removing the watermark from a JPEG isn’t just about erasing the visible text or logo — the AI also has to reconstruct a background region that has been deformed by the compression artifacts the watermark created. The lower the original JPEG quality setting, the worse this problem is.
Not all JPEG watermarks are equally hard to remove. There are two meaningfully different cases:
Text watermarks (like “Shutterstock” or “Getty Images” repeated across a photo) tend to be semi-transparent — they blend the text color with the background below. The AI has more background signal to work with, since the original image shows through. Results are generally good, even on JPEG, as long as the image isn’t heavily compressed.
Logo watermarks (opaque brand logos in corners or center) completely obscure the background underneath them. There is no original pixel data visible — only the logo. The AI has to invent the background from contextual clues. On high-quality JPEG images, this works well. On heavily compressed files, the surrounding artifacts make reconstruction noticeably rougher.
If the image quality is poor, consider these steps before uploading to any removal tool:
Goodbye Watermark
Goodbye Watermark uses a multimodal AI model that understands image context — not just pixel patterns. It reconstructs the background under watermarks on JPEG images, handling compression artifacts intelligently to produce clean, seamless results.
Upload your JPEG, download as PNG. Free for up to 5 images per day. No account, no software install.
Saving the AI output as PNG instead of JPEG has a concrete effect on quality. PNG is a lossless format — every pixel value the AI produced is saved exactly as-is. When you save the same output as JPEG, the codec compresses it again, introducing new blocking and ringing artifacts on top of what the AI reconstructed.
The visible difference is most apparent at the watermark boundary — the edges of the filled region. PNG preserves the clean blend the AI computed. JPEG re-compresses it, often creating a faint halo or blocking pattern that’s especially visible on smooth backgrounds.
The file size will be larger as PNG, but for a single image where quality matters, that’s the right trade-off.
Yes, significantly. High-quality JPEGs (quality 85+) produce results close to PNG. Low-quality JPEGs (quality below 60) have heavy blocking artifacts around the watermark area that the AI has to work through — results are still good but may show subtle texture inconsistencies at the removal boundary.
Most AI watermark tools let you download the output in PNG format regardless of input format. Always choose PNG for the download to avoid re-introducing JPEG compression artifacts.
Full-frame watermarks — like a large diagonal text overlay covering 60%+ of the image — give the AI less original background to work from. Results are less consistent the more of the image is covered. Semi-transparent overlays work better than fully opaque ones in this scenario.
It depends on the source. Removing watermarks from stock images you haven't licensed violates copyright. Removing watermarks from images you own, created yourself, or have appropriate rights to is generally fine. Always verify your rights before removal.
Related articles
Ready to remove a watermark?
Free, no signup required. 5 images per day.