Restore Old Photos With AI: Turn Blurry Family Pictures Into Sharp Memories up to 8K.

What AI Restoration Can and Can't Do

Let's be honest about the limits before getting into the workflow:

AI can reconstruct: Low resolution (adding realistic detail at higher resolution), film grain and noise (smoothing without blurring underlying detail), JPEG compression block artifacts (removing the grid pattern that appears in low-quality saves), and fine facial detail from low-resolution face regions.

AI cannot recover: Information that genuinely isn't there. If part of the image is a solid smear — water damage, physical tearing, severe exposure issues — the AI is reconstructing from almost nothing. Results vary. Motion blur (camera shake) is a specific problem that upscaling doesn't fix — it makes the blur sharper, which can look worse. Use a dedicated deblur tool before upscaling for camera-shake images.

With realistic expectations set: for the most common family photo restoration case — a small print scanned at 300–600 DPI with film grain and moderate fading — AI upscaling produces dramatically improved output.

The Right Model for Each Damage Type

Portraits and photos with faces

Use Bria Increase Resolution. This model is trained specifically for face preservation — it reconstructs skin texture, eye detail, and hair structure at 2x or 4x resolution without changing the subject's appearance. For a 1960s portrait that has been scanned at 300 DPI and shows grain and low resolution, Bria produces the most faithful restoration.

The key difference from general upscalers: Bria doesn't "improve" faces by generating more attractive versions — it reconstructs what should realistically be there based on the surrounding context. Your grandmother still looks like your grandmother.

Group photos and outdoor scenes

Use Real-ESRGAN. It's excellent at handling the combination of film grain, foliage, and architectural detail that appears in outdoor family photos. Credits are priced per upscale, so a whole box of old prints can be restored for the cost of a credit pack.

For the technical background on how Real-ESRGAN handles different types of degradation, the original Real-ESRGAN paper (arXiv:2107.10833) is the authoritative reference.

Black and white photos

Both Bria and Real-ESRGAN handle black and white correctly — they treat the image as grayscale and upscale accordingly. Do not convert to color before upscaling; colorization is a separate step if desired (done after upscaling, with a different tool).

How to Scan Old Photos for Best Restoration Results

The scan quality determines the upscaling ceiling. A poorly done scan can't be fully recovered, even with the best AI model. Here's the scanning protocol:

• Resolution: Minimum 600 DPI for standard prints (4×6, 5×7). For wallet-size or small prints (2×3), scan at 1200 DPI — the higher DPI compensates for the smaller original.
• Format: Save as TIFF or PNG — not JPEG. JPEG compression will introduce artifacts before you even start upscaling.
• Clean the original first: Fingerprints, dust, and hair on the scan surface become defects in the scan. Use a clean microfiber cloth on the print and the scanner glass.
• Scan multiple times if needed: For damaged prints, scan at different exposure settings and composite the best results before upscaling.

Step-by-Step: From Old Print up to 8K

1. Scan at 600 DPI (or higher for small prints).Save as TIFF or high-quality PNG. A 4×6 print at 600 DPI produces a 2400×3600 pixel file.
2. Light clean in a photo editor (optional).Remove obvious dust spots with the healing brush. Adjust levels if the scan is very dark or faded. Don't sharpen — that's the AI's job.
3. Upload to UpRes. Drag the file or paste a URL into the demo.
4. Select model. Bria Increase Resolution for portraits, Real-ESRGAN for group shots and outdoor scenes.
5. Choose 2x scale. A 2400×3600 scan at 2x gives you 4800×7200 — more than enough for large prints or framing.
6. Download and inspect at 100%. Check the critical areas: eyes, hair, edge detail. If a face has been upscaled well, the irises and skin pores should show clear structure.
7. For printing: export at 300 DPI. A 4800×7200 image at 300 DPI prints at 16×24 inches — poster size, frame-quality output.

Batch Restoration: Old Photo Albums

If you're restoring an entire album rather than a single photo, batch processing is significantly faster than one-at-a-time. UpRes Pro supports ZIP batch upload: scan all the prints, ZIP the files, upload once, download a ZIP of all the restored versions.

For albums of 50+ photos, consider the REST API approach described in the batch upscaling guide. For mixed content (some portraits, some group shots), you'll need to segment by content type and run separate batches with Bria vs. Real-ESRGAN.

FAQ