Correction, 2026-10-02: this page offered Clarity as the way to upscale a scan without saying that the 8frame upscale tool starts on Topaz, which always enlarges 4x and is priced by the size of the result (65 credits for a 4 x 6 inch print scanned at 600 dpi, not 11), or that Clarity always enlarges 2x and is diffusion-based, so it can redraw a face. It left out BRIA Increase Resolution (6 credits). Its total of 20 to 30 credits for a photo repaired and upscaled for printing does not hold for a full scan, and Flux Edit's 7 credits holds only for a 1 MP image. It also stated as fact that free tools handle scratches, dust and modest upscaling well, which we had not measured; those lines are now advice.
Photo restoration is the most emotionally loaded thing people do with AI imaging, and the one where restraint matters most. A restored family photograph is not a design asset. It is the only remaining image of someone, and the difference between repairing it and replacing it with a plausible fiction is a difference the family will feel even if they cannot name it.
Start the mechanical repairs with a free tool. The judgement is yours.
TL;DR
- Free tools are the place to start for scratch removal, tear repair, denoising and modest sharpening; check every repair against the scan
- Faces need the most restraint. Face enhancement can drift toward a generic face, which on a relative reads as wrong
- Colourisation is invention, not recovery. The colours are plausible guesses, and should be labelled as such within the family
- On 8frame: the upscale tool starts on Topaz (fixed 4x, priced by output size, 65 credits for a 600 dpi scan of a 4 x 6 inch print); Clarity is 13, always 2x and diffusion-based; BRIA Increase Resolution is 6. Flux Edit about 7 for a 1 MP image, Kontext Max 13. Stills only
- Scan properly first. No restoration step recovers detail that a bad scan never captured
Start with the scan
This is the step that determines everything downstream and the one people skip.
- Use a flatbed scanner if you can, at 600 dpi minimum for a standard print, higher for anything small.
- A phone photo works if the scanner is not available: flat surface, even indirect light, camera parallel to the print, no flash. Flash leaves glare, and a tool can only guess at what the glare hides.
- Scan the damage, do not hide it. Do not use the scanner's own "restore" or "descreen" settings. Capture what is there and make repair decisions deliberately afterwards.
- Keep the original scan untouched, always, and work on copies. This is the one irreversible part of the process.
Where to start with a free tool
Scratch and tear removal. A white scratch across an otherwise intact area is filled in from the surrounding texture, which is the kind of inference these tools are built for. Compare the repair with the scan at full size before you keep it.
Dust and speckle. The same kind of repair; check it the same way.
Denoising film grain. Possible, but see the restraint note below.
Modest upscaling. Scan at 600 dpi or more first, so the upscaler starts from real detail, and enlarge only as far as the print needs.
Contrast and fade correction. A faded photograph has compressed contrast, and expanding it back is a real recovery rather than an invention.
Where restraint matters
Faces. Face enhancement can pull toward a generic face, and the result is a relative who looks slightly like a stranger. Apply the least possible, and evaluate by showing it to someone who knew the person rather than by looking at it yourself.
Grain removal. Film grain is part of what makes an old photograph read as old. Strip it entirely and you get something that looks like a modern digital image of a period scene, which is a different object.
Missing regions. When a corner is torn away entirely, a tool will generate a plausible continuation. That content is fiction. For a family archive, either leave the damage or keep an annotated note that the region was reconstructed.
Colourisation. The colours are statistically likely, not recorded. Her dress was probably not that blue. Colourise if the family wants it, keep the monochrome original, and be clear about which is which.
What it costs on 8frame
Canon prices from the 8frame app; a credit is about $0.01 at pack rate:
| Task | Route | Credits |
|---|---|---|
| Upscale, the option the tool starts on | Topaz Image Upscale, fixed 4x | 11 per started 24 MP block of output, up to 281 |
| Upscale, flat price, always 2x | Clarity (diffusion-based) | 13 |
| Upscale, flat price, 2x or 4x up to 8192 x 8192 | BRIA Increase Resolution | 6 |
| Targeted repair or adjustment | Flux Edit (FLUX.2 pro) | about 7 for a 1 MP image, more for larger ones |
| Second edit model | Kontext Max | 13 |
| Isolate the subject | background removal | 12 |
For a family photo, the choice of upscaler matters more than its price. Clarity is diffusion-based and redraws detail, so compare every face with the scan before you keep the result; BRIA Increase Resolution is built to preserve the original content.
A worked example, computed from 8frame's prices rather than run: a 4 x 6 inch print scanned at 600 dpi is 2400 x 3600 px, about 8.6 MP. Topaz enlarges it 4x to about 138 MP, which is 6 started blocks, so 65 credits. Clarity makes it 4800 x 7200 px for 13. BRIA's 8192 x 8192 ceiling allows 2x at this size, not 4x, for 6. Flux Edit bills input plus output megapixels, about 29 on the full scan, and its output tops out at about 4.2 MP, so repair first and upscale last. On a 1 MP image, one repair pass plus one flat-price upscale is 13 to 26 credits, about $0.13 to $0.26.
For a single precious photo, a free local tool is the right choice and paying is unnecessary. Paid workflows make sense when restoring an entire box.
The ethical line, stated once
Repairing damage restores what was photographed. Generating a face that was never in the frame, opening closed eyes, changing an expression, or adding a person does not restore anything; it creates a new image and presents it as a memory.
Families will accept a scratch. They will not thank you for a grandmother who is subtly not her.
FAQ
What is the best free photo restoration tool? For one-off family work, a free browser tool or an open-source local one. Local matters if the photo is private. Our roundups of free AI upscalers and free AI photo enhancers name tools with the limits stated on their own pages.
Should I colourise old photos? Only if the family wants it, and keep the original. The colours are guesses.
Can AI restore a photo where the face is destroyed? It can generate a face. That is not restoration, and for a family archive it is the wrong call.
Does 8frame do photo restoration? It has the pieces (upscaling, editing, background removal) for stills. It is not a dedicated restoration tool.
Sources
- 8frame prices and settings (Topaz as the starting upscale option, fixed 4x, 11 per started 24 MP block of output, up to 281; Clarity 13, always 2x, diffusion-based; BRIA Increase Resolution 6, 2x or 4x up to 8192 x 8192; Flux Edit billed on input plus output megapixels, output up to about 4.2 MP; Kontext Max 13; background removal 12): the 8frame app, checked 2026-10-02
Repair the damage, keep the original, do not invent a face.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.