How to Restore Old, Blurry, and Damaged Photos with AI Face Restoration

Old family photos degrade in predictable ways: fading, scratches, film grain, low resolution, and yellowing. AI face restoration was largely built to solve exactly this problem — it’s the single most common real-world use case for models like GFPGAN. This guide walks through the process end to end.

Step 1: Scan or Photograph the Original Well

Restoration quality depends heavily on the input. Before uploading anything:

• Scan at the highest resolution your scanner supports (300–600 DPI for prints). If you’re photographing the print instead, use even, diffuse light and avoid glare or shadows across the face.
• Keep the print flat. Curled or bent photos introduce blur and warping that restoration can’t fully correct.
• Crop loosely. Leave some margin around the face rather than cropping tightly — the alignment step in most models works better with context.

Step 2: Choose the Right Restoration Approach for the Damage

Not all degradation is the same, and it changes which model performs best:

Damage Type	What’s Happening	Best-Suited Approach
Fading, low contrast	Chemical degradation of the print over decades	Generative-prior models (GFPGAN) — strong at rebuilding from sparse detail
Scratches, dust, tears	Physical damage to the print surface	Generative-prior models, often combined with inpainting
Motion or focus blur	Camera or subject movement at capture time	Either GFPGAN or CodeFormer perform well
Heavy JPEG compression (old digital photos)	Blocking artifacts from repeated re-saves	Both models handle this well
Very low resolution (small or cropped face)	Limited pixel data to begin with	Generative-prior models — they reconstruct rather than just upscale

See our full CodeFormer vs GFPGAN comparison for a deeper breakdown by photo type.

Step 3: Run the Restoration

Upload the scanned or photographed image to a face restoration tool. Using the free browser-native tool on this site:

1. Upload the photo — processing happens locally, so nothing is sent to a server.
2. The tool detects and aligns every face in the frame automatically, including group photos.
3. Restoration runs in under a second per face on typical hardware.
4. Download the result at up to 1024×1024px per face.

Restore a photo free →

Step 4: Pair with an Upscaler for Large Prints

If you plan to reprint the photo at a larger size, run the restored output through an upscaler (such as Real-ESRGAN) afterward. Face restoration rebuilds facial detail; a general upscaler increases overall resolution for the full frame, including background and clothing. Running restoration first, then upscaling, produces sharper results than upscaling alone.

Step 5: Review Before Reusing or Publishing

AI reconstruction is a best estimate of missing detail, not a certainty — especially on severely damaged photos where very little original information remains. Before using a restored photo for something significant (memorial prints, publications, identification purposes), compare it against any other reference photos of the same person and confirm the family is comfortable with the result.

What to Expect: Realistic Results by Damage Level

• Mild damage (slight blur, minor compression): Restoration is close to seamless — most viewers won’t notice it was processed.
• Moderate damage (visible fading, moderate blur, low resolution): Strong improvement, with fine details like individual hair strands or exact skin texture being AI-reconstructed rather than recovered.
• Severe damage (heavy fading, large tears through the face, extremely low resolution): The model produces a plausible, natural-looking face, but treat fine details as reconstructed rather than historically exact.

See real examples across these categories in our before & after gallery.

Related Guides

• Face Restoration Guide — what face restoration is and how the underlying AI works.
• GFPGAN Face Restoration: Before & After Examples — real results across photo types.
• CodeFormer vs GFPGAN — which model to pick for which kind of damage.
• Best AI Face Restoration Tools & Apps — other options if you need batch processing or desktop software.

Frequently Asked Questions

What is the best way to restore an old photo with AI?

Scan the print at the highest resolution available, upload it to a face restoration tool without cropping too tightly, and use a generative-prior model like GFPGAN — it’s specifically strong at reconstructing detail from severely faded or damaged prints.

Can AI restore a photo where the face is barely visible?

Partially. If enough facial structure survives — eyes, general proportions — a generative-prior model can often produce a plausible reconstruction. If the face is almost entirely destroyed, results should be treated as a best estimate rather than a historically exact reconstruction.

Does restoring an old photo upload it anywhere?

Not if you use a browser-native tool. The GFPGAN tool on this site processes images locally in your browser using WebAssembly — your photo is never uploaded to a server.

Should I upscale before or after restoring a face?

After. Run face restoration first to rebuild facial detail, then apply a general image upscaler if you need a larger final resolution for printing.

Can I restore multiple old photos at once?

Yes — the browser tool on this site processes each photo you upload individually and has no daily limit, so you can work through a family photo archive one image at a time without an account.