I recently put together a new video montage using some seriously vintage footage. And when I say “vintage,” I mean 23-year-old miniDV and webcam gold. Naturally, the actors I used have grown older (and infinitely wiser) since those days and moved on with their lives. The faded footage is all that’s left right?

Enter Artificial Intelligence.

While my future posts will dive into the wild world of creative AI (generating entirely new things), today is dedicated strictly to restorative AI—the digital sorcery of making ancient pixel-mush look like it was shot yesterday.

It wasn’t a totally smooth ride. The montage is full of spectacular hits and a few hilarious misses. When you feed ancient tape into a modern neural network, you run into two main perils:

• Hallucinations: No, the AI hasn’t been eating funny mushrooms. In tech speak, this happens when the AI tries to “guess” what a blurry shape is and confidently invents details that never existed. (Think: a smudged hand suddenly getting six fingers, or a shadowy bush turning into a creepy face.)
• Artifacts: These are glitchy digital leftovers. Imagine random blocky pixels, unnatural flickering, or weird halos floating around people. It’s what happens when the software gets confused by Y2K-era webcam compression.AI Magicx

But when you hit the sweet spot, it works absolute miracles. Software like Topaz Video AI, Aiarty, and the AI features in DaVinci Resolve are incredible at smoothing out grain and rescuing lost details.

AI Magicx

The secret trick? The prompt. You have to tell the AI exactly what to fix and what to leave alone. I actually got an AI to write the perfect restoration instructions for me.

I’ve pasted that exact prompt below. Feel free to steal it for your own time-traveling edits!

Restore and upscale this low-resolution webcam image to true 4K quality. 
Treat this as a forensic restoration, not a creative generation. Preserve the person’s identity, facial structure, expression, age, skin texture, hairline and all visible features exactly as they appear in the source image. 
Recover detail only where there is clear visual evidence. Do not guess or invent missing information. If eye colour is unclear, keep it neutral rather than guessing. If facial hair is uncertain, do not create it. If any detail is ambiguous, preserve the ambiguity rather than hallucinating new features. 
Remove compression artefacts, webcam noise, pixelation and blur while maintaining a natural photographic appearance. Improve sharpness, colour accuracy, contrast and lighting without changing the subject or scene. 
Do not beautify, stylise, age, de-age, smooth skin, alter facial proportions, add facial hair, add makeup, change clothing details, change the background, or introduce any new elements. 
The final result should look exactly like the original photograph captured with a modern professional 4K camera, not a different photograph of the same person.