Redux: Enriching Dungeons & Dragons with Generative Art
Following up on my previous post -
This idea didn't work at all, for several reasons.
Midjourney is very bad at generating consistent characters. Players have a very clear idea of what they think their character should look like. When Midjourney generates something different, it can be hard to reconcile or add in fine detail.
Adding onto that, it's difficult to get Midjourney to generate the same character twice, even using the same prompt/seed/seed images. I think other systems like Stable Diffusion are a little better at this. New updates to Midjourney might have also improved the situation.
Placing a consistent character in a scene doing an action is even more difficult. This is certainly possible with lots of effort and/or other AI systems - see demonflyingfox on YouTube - but it didn't work well with Midjourney, especially not in real-time.
Real-time generation is also pretty slow. I mean, it's fast compared to other AI image systems, on the order of a few seconds. But if I have to stop between every setting change and wait in silence while an image generates, then that could be disruptive.
I didn't actually try to run a campaign like this, but I realized that if Midjourney was in our text chat, players would likely be distracted trying to prompt it instead of actually playing the game.
Overall I think generative AI for creating D&D assets is a great idea, but the technology has a long way to go before it works in real-time.