• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Japanese game developer Level-5 reveals how they use generative AI like Stable Diffusion

Level-5, the Japanese developer known for the Ni no Kuni and Professor Layton games, have revealed that they have been using generative AI to assist game development and promotion. In a slideshow document submitted to a Japanese government study group on intellectual property rights in the AI era, the developer gave examples of how they utilize AI.
20231213-24326-001.jpeg

According to a report by ITmedia, Level-5 demonstrated how they have been using the text-to-image AI model Stable Diffusion. For Megaton Musashi: Wired, set to release next year, they had the AI generate multiple draft layouts for the title screen, which the studio's illustrators then used as reference to create a final title screen that would fit the game's concept. For promotional images, Level-5 are not only using Stable Diffusion to generate initial layout ideas but are also using AI (SwinIR) to sharpen up the image quality.
20231213-24326-002.jpeg

20231213-24326-003.jpeg

Level-5 is also known for its Yo-kai Watch games based on the popular children's anime. They used Stable Diffusion to create 3D images of main character Nathan Adams/Keita Amano. The AI learned from multiple 2D images of the character in past games, outputting a variety of 3D images with different lighting, expressions and backgrounds. The developer uses these as a guide for creating the right atmosphere in the game.

For their soccer RPG Inazuma Eleven: Victory Road, Level-5 used parts of AI images to create sections of the in-game backgrounds, such as large crowds and groups of buildings. They then combined these AI segments with the backgrounds and character illustrations created by artists.
20231213-24326-004.jpeg


20231213-24326-005.jpeg

In addition to image generation, Level-5 is also using other types of AI. The company is using ChatGPT for character creation proposals and the text-to-speech synthesizer software VOICEVOX to generate temporary voices for characters. For game promotion, they are using the programming assistance tool Github Copilot and Genie AI (ChatGPT) to generate code and polish up game descriptions in order to create websites in a more efficient manner.
 

Mephisto40

Member
It worrying how quickly it's becoming a reality that Dev's are just going to be able to run their entire game into an AI and press the "make me a sequel" button
 
AI stuff might not be perfect in each and every task, I guess especially texture work (and voice acting are certainly already on good enough level) but if humans only are needed for the quality checks and minor touch ups, it just makes sense and it will accelerate some steps immensely, without really compromising quality at all. Sucks for most of the worker in those fields though, those who won't become the quality check people.
 

RoboFu

One of the green rats
It worrying how quickly it's becoming a reality that Dev's are just going to be able to run their entire game into an AI and press the "make me a sequel" button

Ever watch star trek the next generation? " Holodeck ... 1842 new york detective / mobster game .. run simulation. "
 

StreetsofBeige

Gold Member
Nothing wrong with AI help IMO.

If a tech tool can help make a product better, faster, cheaper etc... who wouldnt want to try it out? If AI sucks, companies wont bother using it.
 

digdug2

Member
EA have been using it for years to make FIFA games.
Madden too. Remember the game breaking bugs from 2021? They're also in 2022 and 2023 because they pushed the "Sequel with newest roster and minor graphical updates but nothing else, let's charge $70 for it" AI button.
 

Crayon

Member
It worrying how quickly it's becoming a reality that Dev's are just going to be able to run their entire game into an AI and press the "make me a sequel" button

We're going to be able to sit down and say "make me a game" holodeck style.

Edit: beaten by insert code.
 
Last edited:
Top Bottom