Photo Fixing AI: Caught in Less than 4K 📸

The Good, The Bad, The Silly

For those that are new around here, this is weekly newsletter where I highlight new and innovative AI products that are worth exploring.

Hey hey!

Happy Friday! We’re back for another 2024 issue.

In this week’s issue:

  • Product (Experiment) of the Week

  • Other AI Things Happened

  • What I’m Watching


After testing dozens of new AI products this week. Here’s my top pick.

Photo enhancement AI wasn’t ready for the nostalgia-trip I sent it on

We've all been there, haven't we? Rolling up our sleeves, diving headfirst into some kind of experiment, fully expecting to come out on the other side with some kind of tangible result ... or at least something that vaguely resembles our initial hypothesis. But let's be real: not every experiment is a smooth sail. In fact, some take a turn into the realm of the unexpectedly hilarious, leaving us with lasting impressions that are a bit more comedic than obviously intended. This time around, we're taking a little detour to celebrate one of those moments in my AI assessment journey: when things didn't just go wrong—they went comedically wrong.

Because sometimes, the best discoveries are the ones that leave us laughing.

The Photo Fixer rundown: The Good, The Bad, The Silly

So this week I caught myself a bad case of wanderlust nostalgia, while going through my old photos, and it inspired this week’s product features.

I set out to benchmark some solutions to bring my favorite pictures from my 2009 trip to Europe into the 2020’s present.

For your critique, here’s young Sam’s attempt at immortalizing the Mona Lisa taken by an unsteady, untrained teenaged hand armed with a 8 megapixel camera under pretty bad lighting conditions.

The original “Potato Quality” Image from my Trip to Europe - Circa March 2009

I figured snaps of the Mona Lisa might be a great testing benchmark for some of the photo restoration solutions out there on the market.

The Competitors’ Roster:

Searching the best across the net, I came up with a list of the top 4 I’d pit against each other for the job of bringing my bad photos to a modern standard of life:

Some field notes going into this experiment 

  • My expectations were not high since I understood that my use case was a touch particular but the results were… pretty varied.

  •  I understand that video restoration is an extremely difficult technical problem to solve. However the compute power required of the task generally should be reduced or more easily processed thanks to lower resolution images (like my masterpiece, Exhibit A)

  • For the developer-minds in the audience: Seeing a perfect rendition of the Mona Lisa sitting in the results would be a very obvious indicator of an overtuned model, since the picture is in the public domain and could potentially be part of the training set that was used to generate the model. So sneaking in an extra element of the sort would be  a telltale of  the individual solutions’ recall performance in generation.

Results Of This Comparison:

The results ranged frankly from (bigger resolutions + just as blurry) to creative reinterpretations, and some mild flavors of improvements in between.

Since a picture is worth a thousand words I might actually leave you with about 5,000 more words on the topic.

Let’s Enhance: Some noted artefacting, alongside some noticeable blur reduction overall. Generally, the lighting seemed better - if we sidestep the aggressive amount of banding. There was a felt absence from any facial recognition.

Remini: That eye makeup’s on fleek! Remini’s facial recognition is by far the most striking in its tuning, and although it looked to rehumanize DaVinci’s iconic works, it stayed still close enough where clarity could be a noted enhancement. Can’t speak much however to the quality of the renderings of the frame and painting background as it seemed to flatten the image & average out some colors in its quest for upscaling & better lighting.

Sharper even tones on the wall, definitely some banding on the lighting

Cutout Pro: light banding one again, plenty of interesting shake reduction and frame sharpness for the outer edges of the frame! The face was very obviously identified, though the entire background, bricks and all were quite blurred. Something happened to the hands here, because “Oh my, are those helix claws?!”

The helix claws, and the interesting generative definitions of the Mona Lisa’s nose


I really tried to give credit where it’s due, by bringing in photography’s main technology provider: Adobe, but woah the results were absolutely mixed. Given that this is was intended to be a review of the AI features within Adobe Express. I found myself extremely puzzled at the generative AI fill features when asking the model to simply unblur the portrait of the Mona Lisa after selecting the subject of the painting.  I was served this creative (read: 🤣🤣) byproduct:

The instruction was to “Reduce the blur on the Mona Lisa”… 🤷🏽‍♂️

Trailing Thoughts & Verdict

After assessing different flavors of the “Market leading AI correction tools” I can say that the industry solutions as a whole still have quite a way to go for threatening the jobs of more proficient photographers let alone competing with the basic tasks that a direct mathematical formula can perform. See Below my grounds for that last claim. this is what happens when you boost the sharpness slider to 100% on adobe's free online non-AI-powered Express solution.

“AI Isn’t The Whole Way There Yet, I think” - Sam 2024


Some other notable news and product launches from this week

  • Germany has decided to approve the European Union's planned AI Act after reaching a compromise. This legislative framework aims to regulate the development of artificial intelligence within the EU, emphasizing innovation-friendly rules and improvements for small and medium-sized enterprises.

  • This is a short recap on the state of AI regulation in Australia based on new developments. Expect audits for public safety & data security, more transparency around mandatory disclosure, and more. Yes they're even referring to the first article in our list!

  • Generative AI is coming to eDiscovery (a digital step in critical legal & risk management processes) once again! This time it’s from Portland Oregon. Exterro Assist is looking to tool a number of players in the Governance, Risk and Compliance space (GRC for short, but let’s not forget the silent CF - standing for Cybersecurity + Forensics)

  • is joining the AI fray with its very own FreeLawChat. A 24/7 service for conversational research on US law. This is a noted jump in accessibility for Spanish speakers in the country since it’s preloaded with full versions of US law documentation in both Spanish and English.

  • It was just a matter of time! From hurricane insurance in Alaska, to cyberinsurance for a flower shop, a provider in the space has launched an offering specifically to address the risks of AI (LLM bias, hallucinations, IP claims, etc.)

    Raises & Mergers Recap:

  • Interesting angle on the M&A space as AI starts to pervade more and more industries. 2010’s buzzword “big data” is providing excise returns during acquisitions. Essentially, more and more transactions are being motivated in part or in full made by market players looking to plug gaps in their datasets.

  • This is a short Market level recap on what it took to raise funding in 2023, and a few comments on key factors that are very much still at play.

  • Lastly, we’ve got the announcement that Nordic Capital acquired a Canadian Fintech company Zafin, a company dedicated to help automate the construction and pricing of personalized banking products to consumers.


""The present is theirs; the future, for which I really worked, is mine." - Nikola Tesla

This is an in-depth interview with a recent-day thought leader around the definitions of Artificial intelligence, the evolution of technology and much more, Jaron Lanier. This is a long talk, but it covers the history of human considerations of AI, why he considers CHATGPT NOT TO BE TRUE ARTIFICIAL INTELLIGENCE. There’s quite a few interesting takes on various aspects of the AI everything, so I’d recommend a watch. If you want to zoom through it, YouTube is gracious enough to give us a 2x playback speed button 😉.

I first heard of Jaron from his 2018 TED talk on how we needed to remake the internet. That too was an interesting thought piece on the interweave of business, human social development, and the somewhat tragic course of the Internet as we first knew it. (Powerful stuff)

Stay well, and until next week.

-✌🏽 Sam

P.S. Interested in having me give you private feedback about a product that you are building? Send me an email: [email protected]


or to participate.