We tried to make a trailer with AI. Oh boy... (prompts included) - Warnament Dev Update #14
We tried to make a trailer with AI. Oh boy... (prompts included) - Warnament Dev Update #14
Hey rulers!
It’s been an intense month, to say the least, as we have been pushing hard to release Warnament on Android! If you wish to play the game right now, don’t hold yourself back. It’s free to play with an option to support us by buying cosmetic items.
But today, we’ll be telling a different story. Warnament is released in beta mode on Google Play (which means you won’t be seeing user reviews, but we will). Our primary focus is still to hit Early Access on Steam, but the release is a release, so we decided it deserves a proper trailer.
Long story short, we ended up having to answer a question: are AI algorithms of today good enough to generate animations for trailers? The answer is… errrr… Well, more on that below.
We cooperated with Lev Koretski, a video maker whose works you might’ve seen on Steam pages of Black Skylands, Graveyard Keeper, Cartel Tycoon, and other renowned games. You should definitely check out his sample reel, especially if you’re an indie game maker!
Since at Luden.io, we’re big fans of experimenting, we decided to refer to the silicon wisdom of AI models for animation. It was not as easy as we estimated, and only tiny bits of what was generated by the model made their way to the trailer.
A quote from Lev himself and some goofy tryouts is below:
[Working with AI models] became not so much a help but rather a perpetually breaking crutch that we still had to learn to work with.
Along the way, I encountered numerous difficulties. I tried many neural networks available for free (such as Runway, Vidu, Google Veo, Dream Machine by Luma), and all of them operated more or less the same way: you upload 1-2 pictures, and the AI creates animation.
You can write a prompt, but often it doesn't work. Or, you need to go through a bunch of formulations to get the desired result. It feels like the developers of neural networks deliberately limit creators with prompts so that as many attempts are made as possible and the [number of props provided per month in a] subscription runs out.
Here’s a bunch of things we got in response to our prompts. Check out this nightmare fuel (prompt - “finger presses the red button”):
Animating machines worked fairly ok, unless they had rotating elements, such as wheels or propellers, in which case things tended to turn out… not so well (prompt: “Panoramic camera movement, left to right, planes flies shooting at the houses, fire below“):
And how do you like those people morphing into artillery? This takes the concept of giving one’s life for their country to a whole new level:
And here’s the Eiffel Tower merging with a house (prompt: “slow approach to a nuclear explosion”):
Can you spot that one soldier who’s running backwards?
Some more creepy hands to top all that (prompts: “finger presses the red button”, “flags fluttering in the background”):
To be fair, just a couple of years back AI models were drawing creepy pictures, but a lot of progress was made in this field since. So chances are good that one day they will become a trustworthy video maker’s helper. This day just has not yet come.
As for the trailer - let us know how you like it! And if you want to discuss anything Warnament-related with us directly (or just exchange some memes), come by our Discord server.
Just a reminder in case you skipped it: Warnament is now available to play for free on Google Play! Next time, we’ll tell a couple of stories about how we were preparing for the release.
Until we meet again!