Why Old Photos Look Bad (and What AI Actually Fixes)
Old photos fail for three main reasons: low resolution (film grain amplified by cheap scanners), chemical degradation (yellowing, fading), and physical damage. AI upscaling directly solves the first problem and partially addresses the second.
Flare enlarges everyday photos without inventing detail. Lumen recovers more of what was actually in a scan, up to 8×. Mirage is the one that invents plausible texture, and it is the wrong choice for a family archive.
For faces, Visage restores the face and leaves the rest of the frame alone. Atelier runs that after a denoise when the photo looks like a portrait, then finishes with Lumen.
Step 1: Scan at the Right Resolution
AI can only work with what you give it. A 72 DPI smartphone photo of a print will produce worse results than a proper flatbed scan. Here's a practical DPI guide:
| Photo Type | Scan DPI | Output Target |
|---|---|---|
| Standard 4×6 print | 600 DPI | up to 8K upscale |
| 35mm slide or negative | 1200–2400 DPI | 8K upscale |
| Medium format film | 800–1200 DPI | up to 8K upscale |
| Large format (4×5 or 8×10) | 400–600 DPI | up to 8K upscale |
If you're photographing prints with a smartphone rather than scanning, shoot in the highest RAW or JPEG resolution your phone supports, with even lighting (no flash) at a 90-degree angle.
Step 2: Choose the Right Model
UpRes offers multiple models, and the best choice depends on your photo type:
- Flare
Landscapes, buildings, and objects. Fast everyday upscale that does not invent detail.
- Visage
Faces and group shots. Restores faces and leaves the rest of the frame alone.
- Lumen
Print and scans when you want the most recoverable detail, up to 8×.
- Atelier
A full restore: denoise, then faces if the photo looks like a portrait, then Lumen.
Step 3: Pre-process Before Upscaling
If your photo has yellowing or color cast, correct it in any photo editor before uploading. AI upscalers preserve color faithfully — they don't correct hue shifts. A quick auto-levels pass in Lightroom or even Google Photos can dramatically improve the final result.
For photos with physical damage (scratches, dust), clone-stamp the worst blemishes first. AI upscaling will sharpen and enlarge damage just as readily as it sharpens and enlarges detail.
Step 4: Upscale with UpRes
Upload your scan to UpRes, select your model, and choose your target resolution. For archival output, up to 8K is sufficient for most 4×6 prints. Use 8K if you plan to print larger than A3.
Processing takes 5–30 seconds depending on the model. Download the result as JPEG, PNG, or WebP. For archival storage, use PNG (lossless). For sharing digitally, WebP gives the best file size at equivalent quality.
Step 5: Post-process and Print
After upscaling, you may want a final sharpening pass and optional noise reduction in Lightroom. The AI output will typically have slightly more micro-contrast than the original — which looks excellent for prints but may need softening for screen display.
For printing, send the upscaled file to a lab at 300 DPI. A 4K UHD image (3840×2160) prints cleanly at 13×7 inches. An 8K image prints at 26×15 inches — large enough for a framed canvas.
Related
Frequently Asked Questions
Can AI fully restore a damaged old photo?
Flare and Lumen improve sharpness and recover fine detail. Visage restores faces. Severe physical damage — torn edges, ink stains — still needs manual retouching first.
What resolution should I scan old photos before upscaling?
Scan at 600 DPI minimum for a standard 4×6 print. This gives roughly 3600×2400 pixels before upscaling. At 1200 DPI you get 7200×4800 — excellent source material for 8K output.
Which UpRes model is best for old black-and-white photos?
Flare is the fast everyday upscale for archival scenes. Visage restores faces. Lumen recovers print detail up to 8×. Atelier runs denoise, then Visage if the photo looks like a portrait, then Lumen.
Does UpRes colorize old photos?
UpRes focuses on resolution and detail enhancement, not colorization. For colorization, process your photo through a colorization service first, then run it through UpRes to achieve sharp, high-resolution color output.