Can I use AI to remove a watermarked background from a supplier image legally?
3 min read
Quick Answer
Usually no. A watermark is Copyright Management Information under 17 USC Section 1202, and removing it without the rights-holder's permission is an independent violation that carries statutory damages of $2,500 to $25,000 per image under Section 1203(c)(3)(B), on top of the underlying copyright infringement claim. The narrow exception is when the supplier has explicitly given you permission (often in a dropshipping or wholesaler agreement) to use unwatermarked versions, in which case ask them for the clean source file instead of scrubbing the watermarked one.
This is not legal advice
This article is general information about US and EU rules as they apply in 2026. It is not legal advice. For decisions that affect your business, talk to a lawyer licensed in your jurisdiction.
What the law actually targets
Watermark removal sits inside the Digital Millennium Copyright Act, which treats Copyright Management Information (CMI) as a separate protected layer from the underlying image. Two violations stack:
- Copyright infringement for using the underlying image without a license.
- CMI removal under 17 USC Section 1202 for intentionally stripping the mechanism the rights-holder used to protect the work. Statutory damages run from $2,500 to $25,000 per violation under Section 1203(c)(3)(B). Courts can award up to triple damages for repeat violations within three years.
The EU has a parallel rule. Article 7 of the InfoSoc Directive 2001/29/EC requires member states to provide "adequate legal protection" against unauthorised removal of rights-management information, and watermarks are treated as RMI. Penalties vary by member state but the prohibition is consistent across the bloc.
Supplier-permitted use vs. unauthorised scraping
The legal analysis turns on whether the supplier has actually licensed the image to you. The watermark is the red flag, not the verdict.
| Situation | Status |
|---|---|
| Supplier provides watermarked previews and unwatermarked files to verified resellers | Use the unwatermarked file the supplier sends. Removing the watermark from the preview is unnecessary and still a Section 1202 act. |
| Wholesaler dashboard with a "download for resellers" link | Use that link. The license usually accompanies the clean file. |
| Image scraped from a supplier listing or marketplace without a reseller agreement | No license to the image. Removing the watermark adds a Section 1202 claim on top of the infringement. |
| Stock library with watermarked previews | Buy the licensed file. Watermarked comp images are explicitly not for commercial use. |
In practice, if you have to use AI to get rid of the watermark, you almost certainly do not have the rights to use the image. Suppliers who want you to use their photos give you the unwatermarked source.
The safer path: shoot the product you already have
If you have the physical product, the watermark question goes away. A phone photo on a clean surface is enough source material for Nightjar to produce ecommerce-ready imagery.
Nightjar separates the photographic look (camera, lighting, mood) into a reusable Photography Style and the framing, angle, and pose into a Composition, so you can apply the same visual direction to each subsequent product. The Product Listing Image Workflow then renders a clean, controlled output from your own photo, and you can save the full setup as a Recipe (a saved Create-form configuration that captures the style, composition, model choice, background, and output settings) so the next ten products in your catalog look like one shoot rather than ten experiments. Images you generate this way come from your own product photo, which avoids the supplier-watermark problem entirely.
For dropshippers without inventory, the option is to order one sample, photograph it once with a phone, and reuse the resulting Recipe across the catalog. The cost of a single sample is generally lower than the exposure of a Section 1202 demand letter.
Consistent and on brand AI photoshoots, optimized for conversion.
Nightjar