Harker’s Escape Lesson 2: Full Generative AI Gameplay in 3D is Currently Out of Reach

A Year of Discoveries at Atomic Dirt

In Harker’s Escape, you were able to break the legs off of any tables or chairs, type a command like, “Combine the legs to form a cross” or “Make the legs into a crucifix” and voila!

In your hands you’re holding one of Dracula’s least favorite objects in the world.


It was a crude crafting implementation but it made sense for the scenario. It worked well. Dracula’s AI would understand what had changed and react accordingly. If Dracula was in the room and saw your wooden cross, he would get rather angry and demand you put it down. 

Okay, now let’s extend that into a scenario when you can poison Dracula.

It’s a cool solution to figure out how to escape. But it also taught us one of the biggest limitations we have right now in crafting a first-person game: full generative AI gameplay in 3D is currently out of reach. Cutting yourself to get blood, mixing foxglove powder into your blood, etc., worked well because we didn’t really have to “show” these particular objects with 3D models or the actions with animations. These are micro-actions games commonly skip through and players don’t expect to see. Because of that we have commonly accepted video game design solutions here: cutting yourself to get blood -> simple red flashes on the UI, making the poison blood potion -> simple potion SFX.  Just thinking about that makes me want to spam Diablo potions.

But here’s another escape scenario that should work in Harker’s Escape: the player finds a wire, attaches the crucifix to it, and then chokes Dracula. For that we’d need to create new 3D objects and animations. But if we have the AI write on the screen what happens to Dracula as you choke the life out of him, but his model is just standing there looking at you, and then suddenly his death animation kicks in…then what the hell are we building here? A full 3D depiction of this one scenario where the player chokes out Dracula would take hundreds of hours to develop, and for a game like Harker’s Escape to play as intended, the total number of gameplay scenarios would be endless. 

Takeaway: No doubt one day there will be impressive new AI gen technologies that whip up new Unity-ready models and animations in real time, but it is not this day. 

“And it won’t be any day soon either!” Aragorn clumsily continued.

Next on The Dirt: we understand your intentions.

 

 
Previous
Previous

Harker’s Escape Lesson 3: LLM Analysis - Understanding Intent > Response Accuracy

Next
Next

Harker’s Escape Lesson 1: LLMs Don't Automatically Make Gameplay Better