The GMTK thumbnails you don't see (Patreon)

Hello Patrons!

GMTK is quite famous for its video thumbnails. I put a lot of time into making these little rectangles look attractive and click-able. Each one could be a little work of art!

But... the thumbnail you see on YouTube is usually not where the thumb started. Every thumbnail goes through a lot of iteration and different ideas. And sometimes I'll upload a thumbnail - and replace it down the line. Multiple times!

So in this post I wanted to share some work-in-progress thumbnails to show you how these things change over time. And to explain some of my thoughts and philosophies behind making 'em. Hope you enjoy!

First up - GMTK thumbnails weren't always so pretty! For the first year they all stuck to a rather dull template: a piece of art, with a black box over the bottom half, and the video title in a simple font.

I thought it was nice to have them all be uniform - but I soon got bored of that look and so eventually went back and changed them all to be unique thumbs. Two examples, above.

And a couple more examples from those old times.

If you've got some nice key art of an iconic character to work with, it's pretty easy to make a good thumbnail. But it's much harder to make something nice from a game like The Witness, which is essentially just a bunch of grids and lines.

So this Jonathan Blow video initially went live with that hideous pink blob in the top left - but I soon realised I needed to change it. I messed around with a more Braid-inspired thumb, but then came up with the idea to use Jon's silhouette from Indie Game the Movie.

This thumbnail, for a video looking at Deus Ex: Mankind Divided's hub level in Prague, also caused a bit of an issue. Here, I got stuck on this phrase "In Praise of Prague", which seemed like clever alliteration - but ultimately isn't as clicky as a more gamer-centric term like "open world".

Here are two more examples of making the titles a bit more pithy. I think "playing past your mistakes" is a bit more general audience, whereas "failure spectrum" sounds more academic.

And people scan through hundreds of thumbnails when they're on YouTube, so it's easier to grab their attention with a few big words than a wall of text. Crunching "should roguelikes have persistent upgrades" to "roguelikes and progression" worked here - you can always write more in the title.

Here's an example of a thumbnail going through small tweaks and adjustments to get it right, for the video on remaking Metroid 2. Like changing the font, and making the important stuff stand out more. Just like in game design, rapid iteration is the way to go: I rarely use the first draft of a thumbnail - I keep tweaking and remaking it until I'm happy.

It was tricky to make a thumbnail for the video on puzzle design because, like with The Witness above, there aren't always iconic characters (with good art available) in these games. Portal seemed like a good shout so I tried a few takes using the game's iconography. But ultimately settled on using Snakebird, instead.

I often have an idea for a thumbnail in my head, but it doesn't always work when I get into Photoshop. So the concept, of a PS1-era Solid Snake laid over the date 1998, took a while to get working. The one where Snake's gun is poking through the hole in the 9 is named "lolno.psd" - I guess I knew it wasn't going to work!

The video on balancing multiplayer games went through a few iterations. I liked the composition of the first one, where a bunch of games can be seen in a shattered mirror - but it's way too noisy and cramped for a thumbnail. You have to remember that these things are often absolutely tiny when they end up on YouTube. I spend a lot of time zooming far out to see how the thumb looks at, say, 25% size.

The video was also, at one point, going to be split into two parts. How to balance the game during design, and then how to change the balance post release. That didn't happen - which is a shame, simply because I liked the mirrored thumbnails.

Level design thumbnails are always tricky - I struggled to make a good one for the video on Hitman 2's Miami level, and so the video didn't get many views on release.

I liked the first one, which shows an overhead view of the stage (taken by me, using a camera hack) - but, again, it doesn't read well when shrunk down to YouTube. And the second was okay, but the video still was struggling to perform.

I then changed the thumbnail to the final one you see, as a last ditch effort to save the video... and it totally worked! Within a few days the video started racking up views and has since gone on to get a million hits. There's a really clear point on the analytics graph where changing the thumbnail completely altered the video's trajectory.

I now use this as an example when urging people to change the thumbnails of their underperforming videos.

A couple examples of simply trying too hard... and needing to step back and keep it simple. If you've seen any of my thumbnails you'll know that I like it when the character is slightly in front of the text. It gives the thumbnail a bit more depth and visual interest. But here are two examples of taking it too far and hurting the thumbnail's legibility.

I also know that YouTube uses AI to pluck text from the thumbnail and figure out what the video is about. If the robot can't read it, that could potentially hurt its chances with the all-important algorithm.

My monster video on Mass Effect went through a few changes - at one point it was going to be split into multiple parts, so the second video would have had a red thumb featuring some of Shepard's squad. But in the end it became a single video. So Garrus was gone.

I also went back and forth on having the male Shepard be in the thumbnail (there was also one that just had male Shepard, but I can't find that file). He's slightly more iconic and potentially would lead to more clicks - but I think Femshep looks cooler (her red hair contrasts nicely against the blue background) and I try to have a nice mix of male and female characters on my thumbs.

You should be able to scroll down GMTK's archive and see a nice mix of different heroes.

Some videos can't be saved, no matter the thumbnail. This Design Icons video, on the history of Britain's bedroom coder scene, is probably too niche to ever gain a big audience. Still, I spent plenty of time trying to get the thumbnail right.

Originally the episode was supposed to focus on Elite, so that first thumb is from that earlier iteration. The second is supposed to evoke old Spectrum title screens, with the eye-searing colour palette available to ZX coders. The third features the computer itself. And the final thumb is a simple one - but the background is actually the Commodore 64 boot screen.

The JPRG video did a lot better - and again went through many versions. It was supposed to focus on Ultima and Wizardry, so the first thumb reflects that. Then it became a more general video on the history of JRPGs. This is one of the few times where I paid for some stock photography for a thumb - I found a nice pile of dice on some stock site and hard to fork over a few quid to get it.

The text was a little hard to read at small scale, so I changed it to a Dragon Quest-inspired image. But ultimately I missed the dice (and I paid for that pic, dammit! I'm gonna use it!), so the final thumbnail is a mix of the two.

When I was first starting Design Icons, I made a bunch of thumbnails for potential episodes. These won't get used now (even if I make these videos, the titles and thumbs will be different), so I'll leave them here for you to look at!

Speaking of unused thumbnails... here are four videos I never made! A Boss Keys episode on the GBA Castlevania games. A video series about the design evolution of different characters (including Simon Belmont). A video on the concept of core pillars. And one on the ending of Mass Effect 2. Maybe one day!

I liked the idea of having a map of the DOOM Eternal arena in the thumbnail. But, once again, it's too small to see properly. I'm still not happy with this thumbnail, but sometimes you've just go to let it go.

Sometimes the image is pretty easy to pick - but the getting the text right is harder. "Two type of motivation" seemed liked a smart idea, considering the success of "Two types of random" - but it didn't quite capture the point of the video. 

And at one point the video was going to be called something like "why Switch doesn't have achievements", but that felt too much like clickbait (That's only mentioned at the very end). So, I just went with something else entirely!

School of Stealth was originally to be called "What makes a good stealth game" - before I realised that that's an almost impossible question to answer. In the end it became a three part series with thumbnails featuring Sam Fischer, Big Boss, and Batman. Here are some unused variants, with characters from Dishonored, Mark of the Ninja, and MGS 5.

I have no idea why this Shovel Knight video went through so many iterations. I think I got a bit obsessed. Maybe it was the struggle to fit four different characters on a single thumbnail - it was never going to work, so every version felt messy and cramped. Sometimes (almost always, in thumbnail making) simpler is better.

Oh god, now I just look totally obsessed. This was another hard video to capture in a thumbnail. I initially wanted to show the video's concept - the early greybox prototype art, or the fact I was talking to the developer. But that didn't quite hit - so went with something more simple. Eventually it retroactively became an On the Level video, and that went through a fair few iterations as well. Let it go, Mark! Let it go!

See - only one major iteration each! I'm not completely bonkers!

Once again, iteration is important. Sometimes the idea works, but it just needs a bit of tweaking to look right.

When a video first goes live, it's being shown to your biggest fans: your subscribers. So I can use the title and thumbnail to speak to them: in this case, hey! Boss Keys is back! Over time, as the video gets recommended to a wider audience, I may tweak it to be more generally applicable. In this case, switching it to the name of the game.

Similarly, it's tempting to make videos in a format all look the same, and carry the same branding. To make the channel look neat and tidy. But ultimately that's not really how it works on YouTube: someone will see your video for the first time and may have no idea about "GMTK" or "On the Level". So you just need to sell them on that specific video.

I think I found a happy medium in On the Level, though: they just say "Level Design", but all with the same composition - so they still look part of a wider group.

A few years back, I realised that it would be nice to show a Game Jam winner on the thumbnail itself, to highlight the awesome games that get made for the jam. It's hard to know what will look best, so I usually make a couple and swap between them until I find the one people click on most.

Hey YouTube, when are you gonna implement A/B testing for thumbnails?

Sometimes the idea in your head just doesn't work. I wanted to feature the big button prompts you get during a moral choice... but it looked messy and the Little Sister's face looks so weird. In the end... say it with me! Simpler is better!

And finally - a recent example, from the Unity tutorial.

Hope you found that interesting! I want to do more behind the scenes stuff in 2023 - it can feel a little self-indulgent, but if anyone cares about the work that goes into GMTK, it's the people who are interested enough to fund its future existence. That's you!

Speak soon

Mark