Disclaimer

This Disclaimer outlines the limitations, appropriate use, and liabilities related to the GFPGAN AI face restoration service provided at gfpgaan.com. By using this tool, you acknowledge and accept the conditions described below.

1. AI-Generated Output — No Accuracy Guarantee

GFPGAN is an artificial intelligence model that reconstructs facial detail in degraded images. While it achieves high accuracy on standard benchmarks (98.2% FaceNet cosine similarity), AI-generated outputs are inherently probabilistic and may not perfectly replicate the original subject in all cases.

Restored images produced by GFPGAN are AI-reconstructed approximations, not verified photographic records. Specific features such as eye color, fine skin texture, hair color gradients, or subtle expression details may differ from the original, especially when:

• The source image is extremely low resolution (below 40×40 pixels for the face region)
• The face is heavily occluded by hands, hats, hair, or other objects
• The subject is photographed at extreme angles (profile, upward/downward tilt beyond 45°)
• The source image contains severe distortion, artistic filters, or non-photographic degradation

We make no warranty that GFPGAN output accurately represents any specific individual. Do not use GFPGAN output as evidence, official documentation, or identity verification in any context.

2. Not a Professional Service

GFPGAN is a free, general-purpose AI tool. It is not a professional forensic, archival, medical, or legal image processing service. Outputs are suitable for personal use, creative projects, and general photo restoration — they are not intended to serve as:

• Forensic evidence or exhibit material in legal proceedings
• Identity verification or authentication for official purposes
• Medical imaging or diagnostic enhancement
• Biometric data for security systems
• Official archival records for historical or governmental use

If you require professional-grade, accuracy-verified image restoration, consult a qualified specialist.

3. Responsible Use of Restored Images

You are solely responsible for how you use the images you restore. Before sharing, publishing, or distributing any restored image, you must:

• Obtain consent from the person depicted, where required by applicable law or platform policy
• Disclose AI enhancement clearly where context requires transparency (journalism, social media, official records)
• Avoid misrepresentation — do not present AI-restored images as unaltered original photographs
• Respect third-party rights — do not restore and republish copyrighted images without authorization

Creating or distributing non-consensual intimate images, deepfakes designed to deceive, or restored images intended to harass or defame any person is strictly prohibited under our Terms of Service and may violate applicable law.

4. Identity and Likeness

GFPGAN uses a generative facial prior (StyleGAN2) to reconstruct detail. In cases of severe degradation, the model makes its best reconstruction based on learned face statistics — not the specific individual's actual appearance. This means:

• The output is a plausible face reconstruction, not a forensically verified likeness
• On severely degraded inputs, the output should be treated as an AI-assisted approximation
• Identity verification based on GFPGAN output alone is not reliable or recommended

5. Third-Party Model and Content Disclaimer

The underlying GFPGAN model is open-source software developed by Xintao Wang, Yu Li, Honglun Zhang, and Ying Shan at Tencent ARC, and is licensed under the MIT License. We did not develop the original model. This site provides a browser-based implementation of the model for public access. We make no claims about the performance of the original model beyond what the authors have published. References to benchmark figures (identity retention, FaceNet scores) are sourced from the original CVPR 2021 paper.

6. Blog and Informational Content

Blog articles, comparisons, and guides published on this site are for informational purposes only. Benchmark numbers, model comparisons, and performance claims are based on publicly available research and our testing at the time of writing. AI models evolve rapidly — figures may become outdated. Do not rely solely on our blog content for technical decisions in production environments. We recommend consulting the original research papers for authoritative performance data.

7. No Liability for Outputs

We are not liable for any outcomes arising from your use of GFPGAN-restored images, including but not limited to: reputational damage, misidentification, legal consequences, or emotional distress. The tool is provided as-is for personal and creative use. All use is at your own risk and responsibility.

8. External Links

This site links to external resources including GitHub repositories, academic papers, and third-party tools. These links are provided for reference only. We do not control external sites and are not responsible for their accuracy, availability, or content. Linking to an external resource does not constitute endorsement.

9. Contact

If you have questions about this Disclaimer or encounter an issue with our service, please contact us at hello@gfpgaan.com or through our Contact page.