If you've been on the internet in the past month, chances are you've seen a remarkably lifelike portrait of Morgan Freeman painted entirely with the iPad app Procreate. It's so unbelievably similar to the photo that inspired it that many have, quite literally, been unable to believe their eyes. But is there any credence to the deniers' claims? Or is the artist being unfairly lambasted for no worse crime than being too good? Well—it's complicated.
It doesn't take more than a cursory glance at Kyle Lambert's website to see that he's a truly talented artist. His work ranges in style from cartoon to sketching to, importantly, photorealism. He's also got a bit of internet notoriety; his YouTube channel has over 50,000 subscribers, and his work is regularly featured on Apple enthusiast sites.
Nothing he's done, however, has come close to the Morgan Freeman portrait, in terms of both quality and notoriety. According to Lambert, the piece took about 285,000 brush strokes and over 200 hours of painstaking work to create. It's also the first major work he's uploaded using Procreate, an app that better enabled him to achieve the near-photographic quality in his portrait of Morgan Freeman. The result, says the artist, naturally looks a little different than some of his more whimsical works.
As soon as Lambert uploaded the video, the internet went berserk; it's got well over 11 million plays on YouTube in just over a month, and has by now appeared on what seems like every major online aggregator. As with most explosively popular internet events, the resounding acclaim was followed closely by what seemed like pretty sound criticism.
The painting was too hyperrealistic, was the main claim. The photo Lambert had used as inspiration quickly surfaced. More technical takedowns followed. Accusations and defences began to ping and swirl, until it was nearly impossibly to keep track of the many ways in which the video above—that only a few days earlier had been the toast of the internet—was a fraud. Through it all, Lambert has maintained his integrity.
Ok. So who do we believe?
The first comprehensive claims of fakery first came to our attention in the form of a set of blog posts on the site Sebastian's Drawings, published three days after the video had caught mainstream attention on sites like this one. In the first entry, Sebastian notes that, when placing the original Morgan Freeman photograph over the purported painting, every single pixel appears to line up absolutely perfectly. That leaves him with only one conclusion:
The only way this could be done, is if you start with a photo and then paint over it, blurring out details. And then replay the video backwards.
Twitter user @RobNonStop even created a helpful infographic to illustrate that very same point.
And it's true; the process in the two videos does appear to be different. It's worth noting, however, that much of the disparity could be a simple byproduct of the Rihanna drawing having been made in Photoshop a Wacom tablet, instead of Procreate on an iPad.
To further silence the skeptics, Procreate decided to do a little detective work of its own. Lambert is one of the app's beta testers; it would be bad business to endorse his work without being absolutely certain that it was legitimate.
Lambert readily agreed to the company's request for the painting's source, and its experts set about examining the raw data for any sign of fraud. Describing the process to Gizmodo, Procreate spokesman Andrew Towns explains:
See, the whole time Procreate was recording everything. Every action in Procreate is saved to the file as soon as the user lifts their finger from the screen. Stroke and actions are saved at roughly 1/30th of a second into two buffers. First they're thrown into the layer composite buffer, and then they're thrown into a video buffer which creates a video frame. So one action = 1 video frame.
Over time, these frames gradually build a video file. When Procreate is backgrounded, the video buffer is ended, creating a video segment. Days of painting will yield multiple, sequential video segments inside the .procreate file. When the user exports their video, Procreate concatenates these hidden segments into a single video file and composites the final frame at the start of the video.
In other words, Procreate was watching the whole time. If it was a fake, the file would tell them so at a glance.
As it turns out, Lambert's .procreate source file contained all of the correct sequential video segments. The company shared the below screenshot with Gizmodo as an example of what they were looking for:
What's more, Procreate was then able to successfully export a complete video using the file inside of the app. The result was a two and a half hour-long video that "clearly showed he really did work on this for more than 200 hours, without importing any images."
So, there you have it! Problem solved, right? Actually, not quite.
We sent the screenshot Procreate provided to Brian McFeeley, a software engineer working in distributed systems processing and API development. He dismissed it immediately.
McFeeley was able to show us how, using a simple code on a Mac terminal, he was able to reproduce the same effect without doing any actual work.
And while it seems ridiculous for a company to go through such lengths to perpetuate a lie, the fact remains that a screenshot certainly isn't definitive proof.
There were other outlying issues, as well. The hardcore painting deniers took to the photo data analysis site FotoForensics in their quest to put all this painting talk to rest. In a followup post spurred by Procreate's endorsement, Sebastian notes that the document ancestors include the same ID as that of the original Morgan Freeman photo by Scott Gries, taking this to mean that "the photo was pasted into the document at some point."
That might not seem like a huge deal in and of itself, but the presence of a photo layer would greatly diminish Lambert's supposed accomplishment. It also directly contradicts what Lambert had explicitly told Gizmodo Español about his process:
At no stage was the original photograph on my iPad or inside the Procreate app. Procreate documents the entire painting process, so even if I wanted to import a photo layer it would have shown in the video export from the app.
Compounding that, the FotoForensics software history would also seem to show that the image was "created entirely on Photoshop CS5 and CS6 on a Mac," not Procreate as claimed.
This might seem like a lot of trouble to go to just to prove someone wrong, but Sebastian notes in his original post that he's embarking on this mission this because claiming digital trickery as artistic achievement diminishes the true artists out there. It's a fair point. Or would be, anyway, if Lambert really were a charlatan.
As it turns out, he's almost certainly not.
It's easy to forget, but the Morgan Freeman painting floating around the internet isn't the original. It's a highly condensed, bandwidth-friendly version of the original, which weighs in at a shopping 4096 x 3072 pixels. To really see what they were dealing with, the Procreate team compared the full painting in all its many-pixeled glory:
So we also exported a PNG file from Procreate at native res, opened it in Photoshop and overlaid the original photo by Scott Gries. Examining the images at full resolution shows both proportion and colour are not a perfect match. When zoomed in it was also very easy to spot obvious brush marks and stroke imperfections. Contrary to the accusations, the actual evidence shows the artwork is not a perfect pixel for pixel image.
But what about the FotoForensics data that seemed to claim the image had never so much as seen the inside of an iPad? Procreate provided a pretty logical explanation for that as well:
These guys were using a .jpg from the web, not the source file. Unfortunately for them, any claims of proof they make from a web image is more than likely going to be wrong.
The original Procreate artwork is 4096 x 3072 px and the press image is 1000 x 740 px. Which means they were examining an image downscaled in Photoshop for a press template (Procreate can't resize images). When images are resized and saved in Photoshop, the metadata reflects such a change. So without the source file, the whole FotoForensics thing was kind of pointless.
Lastly the XMP data showed at some stage the photo by Scott Gries was copied into the press image. Kyle pointed out, that in his Photoshop press template he had "overlaid the original photo to see how close" he got it.
The important part here is that the original photo was copied into Photoshop over the final downscaled, press template. Not into Procreate. Had Kyle imported the original photo into Procreate as a reference, I would still have said the artwork is amazing. The fact the video data inside his Procreate source file show he didn't import a thing, is testimony to this guy's level of skill.
There are holes to this argument—McFeeley points out that the workflow Procreate describes isn't particularly intuitive—but they're mitigated by the simple fact that this is a long way for a company to go to to cover up a situation it never needed embrace in the first place. And interestingly, when McFeeley overlaid Lambert's painting with the original photo, the similarities were not quite as present as the post on Sebastian's Drawings seemed to imply:
The hairs on his face are nearly—but not quite—pixel perfect. It is entirely possible to have achieved this by painting incredibly zoomed in (think Chuck Close paintings) and working in tiny increments, but it boggles my mind
that somebody could have the patience to do that.
In other words, it almost would have been even more trouble to fake it. In addition to not being entirely pixel-perfect, the lines in the purported painting are much stronger and more opaque—something that's certainly indicative of a touchscreen-drawing experience. And while this work certainly seems preposterously tedious, in speaking to Gizmodo, Lambert seemed rightfully proud (if not somewhat bashful) about the sheer amount of time he claims to have put in:
I worked on it over the course of about a month in total. I said on the site that it was 200 hours of work, but it was probably quite a bit more. And that's part of why its really frustrating. You're really excited about this tedious thing you've been working on for so long, and then you have all these people just blowing it off.
Putting those hundreds of man hours together actually produced a video about two and half hours in length, but obviously, no one on the internet is going to have the patience to download something like that, much less watch it. With all that work sped up to under four minutes, it's only natural that the move from blobs of colour to such extreme photorealism seems impossible to have been done by human hands. But according to Lambert, in the full two and a half hour version, "The first ten minutes of the video is really the most important part. Everything after that is just very fine detailing."
Meaning that of those supposed 200+ hours, the majority of it was spent crafting tiny line by tiny line. Mind-numbingly tedious? Certainly. But impossible? Hardly.
Ultimately only Lambert will ever know for sure if his Morgan Freeman painting is the real deal. But it takes a journal of incredible conspiracy theories to think that this is all an elaborate ruse. This is one case where the skeptics, though well intentioned, are fighting a losing battle.
The lesson in all this? Sure, you shouldn't necessarily take claims at face value—but that means questioning the questioners, too.