My mom texted me a photo last week of the Pope in a big white puffer coat, looking like he was about to hit the slopes at Aspen, and she wanted to know if it was real. It wasn't, obviously, but I'll admit it took me a solid three seconds of squinting before I was sure. That's the part that's been rattling around my head for days.
If you missed it, the image (and there were actually a few variations floating around) started making the rounds on Reddit and then Twitter about a week ago, and by the weekend it was everywhere, including in my family group chat, which is usually reserved for birthday reminders and the occasional blurry photo of my nephew. Someone made it with Midjourney, and it got picked up and reposted so many times that half the people sharing it had no idea it wasn't a real Getty Images photo.
Here's the thing that actually matters to me as someone who pays $10 a month for the basic Midjourney plan and spends an embarrassing amount of time in their Discord: this would not have worked two months ago. Midjourney shipped version 5 back on March 15th, and the jump in quality is not subtle. Hands look like hands now, mostly. Fabric folds correctly. Lighting is consistent across a face instead of looking like three different photos stitched together. I've generated maybe 40 or 50 images since v5 dropped and the failure rate on stuff like "person wearing a coat, outdoors, natural light" has gone from "constant" to "occasional." That's the whole story here. It's not that someone got clever with Photoshop, it's that the tool got good enough that being clever wasn't required.
I keep thinking about the fact that the image worked specifically because it was low stakes and a little funny. Nobody's political worldview hinges on whether the Pope owns a puffer jacket. But the mechanism is identical to the one that will eventually get used for something that isn't funny, and I don't think we've built up the reflexes for it yet. My mom is not a dumb person. She's sharper than me about most things. She just hasn't had a reason before now to look at a photo and ask "wait, was this made by a computer," because until about two weeks ago that question rarely applied to a photo of an old man standing on a sidewalk.
There's a smaller, dumber complaint buried in here too, which is that the Midjourney Discord server has gotten borderline unusable if you're not paying for the higher tiers. I'm on the Basic plan, 200 generations a month, and half the time the "newbies" channel is scrolling so fast I can't even find my own prompt results before they vanish off the bottom. I get why they gate the good stuff behind the $30 Standard plan, but sitting in a queue behind a thousand strangers to render a picture of a dog in a spacesuit is not exactly the streamlined creative future I was promised. Somebody's going to build a version of this with a real interface and a real gallery and it is going to eat Midjourney's lunch the moment it ships, mark my words.
Separately, and I'm not going to spend a lot of words on this because I think every tech site on the internet has already said it six different ways this month, but the open letter a bunch of researchers and tech names put out a week or so back asking for a six month pause on training anything bigger than GPT-4 is still being argued about everywhere I look. I don't have a hot take on the letter itself. What I do have is a slightly uneasy feeling that reads a lot like the one I got looking at that Pope photo: the sense that the gap between "impressive demo" and "thing that changes how a regular person interacts with reality" just got a lot shorter than it used to be, and none of us really voted on that pace.
Anyway. If you're going to make images with Midjourney v5, do yourself a favor and add "natural skin texture" to your prompts, it cuts down on that waxy over-smoothed look by a noticeable amount. Small tip, free of charge, more useful than anything I said about the Pope.