Stepping up the AI art game

Maintaining an art style with consistent design cues does not play well with the current image generations toolsets of today. The world of Montavera is quirky, mysterious, and charming (and soon with future properties, foreboding and dark). Finding a way to manage this consistency has been a long journey as AI generation rapidly develops before our eyes. With each new property release we want to make sure we are pushing the technology as far as we can so that what we have in our imagination can make it into the final image.

The most time consuming part of developing the images for the Montavera properties is a battle between rooting out stable diffusion results with negative prompts and a healthy diet of constant image model training repeated over…and over… and over again. A big part of this process is tapping our feet waiting for our computers to deliver the next result as we relentlessly punish our 3090 RTXs building the final 12k Tivea location. Having Theta introduce EdgeCloud and the hosted GPU resources sharing this summer will be an absolute GAME CHANGER. I can’t tell you how excited we are to access the computing horsepower available of an A100 right at our fingertips on the upcoming EdgeCloud update. Like, this is incredible.

Anyway, the Tivea release marks the completion of the Southern part of Montavera from the Long River and below including “The Streets”. I want to share some of the novel techniques we used in creating the Tivea tree-village location. Not only are they individual NFT images, but the properties are part of one single super high-resolution image. This way the properties are both individualized and now share a cohesive single location in the dramatic magical tree village. To show this off the video above will zoom in on each property in detailed full resolution from the final 12k image.

To accomplish this task we invested a lot of time in mastering Control Net, creating our own Lora models based on previous custom art training, and embedding them within a 2nd Architectural model we trained to our tastes. Most of the work was jumping between invokeAI and Automatic1111 to retopologize the low resolution concepts, then running a script to cut sections of the Tivea Tree for “image to image” processing. The up-res process took awhile to master so that the interpretive upscaling perfectly matched across all sections of the Tree. That is, up to the best it could get until we moved everything into Photoshop for final clean up and tone-matching. 

Congratulations to the Tivea property owners! Your NFTs will be minted soon and arrive in your wallets!

Next up in the blog:

1. The rental smart contract has been completed and tested, expect a demo of the functionality and how your dreams of being a Landlord in Montavera will finally come to fruition.
2. Clan Wars has seen an incredibly successful playtest series with seasoned gamers in the physical trials of the game. Look forward to a fat blog entry on that as well and the upcoming plans!