Making the right characters for [insert anything]

by Daniel Abensour

At Hidden Door, we’re building a library of tools to help us bring any story to life, from text to art. As the art director, it’s my mission to make sure we strike an adequate balance between meeting players’ expectations and delivering surprises—while building on certain narrative gaming traditions to prepare our audience for a novel but familiar play experience.

I try to evolve the visual experience towards something that works within the same constraints as visual novels, providing the appropriate visual orientation to assist the text-heavy gameplay in the most frictionless way. This direction becomes slightly more complicated when you have to think about the systems in place to represent everything and anything—but setting up a solid foundation is what we’re doing, a nice little team of art skeletons we can dress up in different ways, skeletons labeled characters, backgrounds, reactions, animations, etc.

A very important part of making a character is designing hairstyles! This is a portion of what we want to make—but as long as the different parts of a character can virtually make anything you need in any given setting… we should be good!

This is where many people would think about using AI—and we do, to an extent. The main advantage of using generative art is being able to do a lot very fast. Quality is a far away consideration if you expect to give a product enough flavor and originality to stand on its own legs.

The quest to get the right balance went through many different versions… most are not illustrated in the above image!

Our characters aren’t drawn generatively because control, style, precision, and more matter in their depictions. Thus, for characters, we’re using a modular art system that assembles puppets of human-illustrated assets (by the infinitely talented Alessandra Criseo) to make unique, memorable characters—while giving players a high degree of flexibility and control. The more assets we make, the more deeply we invest in a library that can make any character (although it will take more illustration work before we have all we need!).

We try to strike a good balance of flavor and usability so assets aren’t restricted to a single environment. For example, most of these outfits used for the PCs in The Wizard of Oz can find homes in many other places.

Today, we’re sharing our new character art style. Early on, we made a neutral-friendly system that echoes avatar-makers. As we evolved, we felt there was an opportunity to bring more value with the art and invested in making unique visuals that could appeal to a more critical audience.

An old mapmaker facing their brand new version.

A little team of new characters featuring the human blep mouth asset.

The background visuals are where we can use generative art.

Some background tests show how we might treat a background to give a sense of location to the players without overplaying our generative hand.

The art style has also evolved to get more stylized in how colors are treated—which makes the game look more unique and hides that generative art can look underwhelming. Using our own refined models and prompts can get us a good way in the right direction, but treating the images with a special post-spice can remove the need for players to read too much into the background art (Can I interact with it? Is it meant to be a 1:1 representation of the world where I can go around and explore every bit of the image?). It brings the right balance of orientation and agency without hinting at more than that.

There are many other pieces to what we’re doing visually with Hidden Door (a library of reactions for how to animate programmatically simple movements for characters, transitions, and interface elements, for example), but I hope you enjoy this glimpse into our perspective on the ideal combination of elements to make our vision possible.