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.
Free tools handle the mechanical parts well. The judgement is yours.
TL;DR
- Free tools are genuinely good at scratch removal, tear repair, denoising and modest sharpening
- Faces need the most restraint. Face enhancement regresses toward an average 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: Clarity upscale at 13 credits, Flux Edit at 7, Kontext Max at 13 for heavier repair. Stills only
- Scan properly first. A good scan of a damaged photo beats any restoration of a bad scan
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 produces glare that no tool removes cleanly.
- 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.
What free tools do well
Scratch and tear removal. Genuinely solved. A white scratch across an otherwise intact area is repaired by looking at the surrounding texture, and that is exactly the kind of inference that works.
Dust and speckle. Same category, same reliability.
Denoising film grain. Works, but see the restraint note below.
Modest upscaling. A small print scanned at 600 dpi and upscaled holds up well.
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 models pull toward an average, 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, a credit is $0.01 at pack rate:
| Task | Route | Credits |
|---|---|---|
| Upscale the scan | Clarity image upscale | 13 |
| Print-grade upscale, per 24MP block | Topaz Image Upscale | 11 |
| Targeted repair or adjustment | Flux Edit (FLUX.2 pro) | 7 |
| Heavier editing pass | Kontext Max | 13 |
| Isolate the subject | background removal | 12 |
A restored photograph, repaired and upscaled for printing, runs about 20 to 30 credits, roughly a quarter.
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.
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.
Repair the damage, keep the original, do not invent a face. The 8frame canvas is free and unlimited, and generation is paid from $19/month.