Responsible AI

How our tools work

On-device tools use classical image processing, such as high-quality resizing, sharpening, contrast, noise reduction, and tonal correction. They improve what is already in the photo and do not generate new content.

Cloud tools use open-source AI models selected for quality and for licences that permit commercial use: Real-ESRGAN for enhancement and upscaling, Swin2SR for denoising, unblurring, and text, GFPGAN for portraits, DDColor for colorization, BiRefNet for background removal, and LaMa for object removal. Low-light and glare correction use classical processing.

When AI adds detail

Some tools can add detail that was not in the original. Portrait enhancement can change facial features, colorization estimates colors, and object removal fills the selected area with generated content. The result screen tells you when a tool may have added detail, so you can decide how to use the image.

Models we choose not to use

We do not use models whose licences prohibit commercial use, including CodeFormer, SUPIR, and BRIA RMBG-2.0.

Reading your result

The fidelity slider controls how strongly a correction is applied. After processing, we show how much of the image changed. A higher number means the tool worked harder; it does not mean new information was recovered. We recommend viewing results at full size before publishing them.

Your photos and our models

Your photos are used only to produce your result. They are never used to train AI models, and cloud files are deleted automatically within 24 hours.

Content Credentials

Results are not currently signed with Content Credentials (C2PA). We plan to add provenance information that records how an image was edited.

Feedback

We welcome questions and feedback about how we use AI at support@enhancerpro.ai.