Conceptualizing the land of Erathe

As I was writing The Sword of the Watch, I thought of scenes visually. That is probably the result of years converting a client’s idea into something you can actually stand in front of with a camera. Most people do not think that way. You have to, or you arrive on location without a clue.

I wanted the series to consist of locations, people, and props you could build — not only look at. Costume and relic had to survive production and, later, a physical object someone could hold.

When tools started letting me simply feed my book text, a few pointers, and then they created some amazing stuff, I began using AI for concepting. Wow, was I ready. The early stuff was soul crushing: Iteration after iteration - it seemed no matter what you did, the picture was just not right. Endless prompting and most results had the same fault as a lot of concept art: cool, with absolutely no regard for actually building the thing.

But things got better. In some ways, a LOT better.

Honestly, it let us generate a lot of concepts — some fabulous, some unusable. Over time, I discovered a new mix of tools for the development pipeline and everything worked more like a toolbox full of really good tools. ControlNet came on the scene, and hit-and-miss iterations were once again greatly reduced. Every update seems to move this technology closer to allowing small shops to generate big ideas and the imagery to match. I still believe real actors (at least for now) will always outperform AI, but many of the traditional methods for compositing, color, animating, creatures - many of these tools are going to change, and are changing.

The test for the crazy new concept generators is same as it's always been for teams of concept artists: does this person, place or object belong next to the books and the Archive, and can the shop build it?

Work  ·  3D gallery