DARK SOULS Ai 3D RINGS

▫️Project: Dark Souls Rings. Recreate 3D rings from Dark Souls using their 2D source art.

▫️Experiment: How accurately can Google’s Nano Banana and Midjourney translate flat 2D image sources into faux 3D models?

▫️Tools: Google Nano Banana (Image to Image), Midjourney (Image to Video), Photoshop (graphic design), DaVinci (editing). 6 clips total. 

▫️Vision: 6 rings from Dark Souls 1 slowly rotating on a dais with their names & a snippet of their real item descriptions. The rings were NOT to be presented at the same angle as the source, but to be faithfully recreated in a demonstrably new way. 

▫️Expectations: I’ve seen how powerful Midjourney and Nano Banana are at taking an existing 2D image and best-guessing the remaining dimensions. While I was unclear on *how* I could best translate a 2D image to 3D, I was confident that the simple scope of my test would yield straightforward results. Easy-in, Easy-out. 

▫️What actually happened: Well it ain’t no easy grab! The original plan was to have the rings revolve 360 degrees. But Midjourney “splits the difference” and created two ring faces and simply rotated the rings 180 degrees. So many Fast Hours were lost on an Ai’s decision-making process that could not be overridden by any prompting. Eventually I settled on slow rotations, but a lesson was learned: Midjourney will compromise your Prompt to create the perfect loop *in a 5 second clip* no matter what.

▫️What I didn’t expect: While it wasn’t easy to recreate some of the more elaborate rings (get outta here Covetous Gold Serpent Ring!), I didn’t expect Nano Banana to recreate them so faithfully. Even the parts of the ring that were hidden are interpreted with surprising confidence. I thought it would be much harder, but with a solid Google to Midjourney pipeline, it’s pretty easy to push these out. I’m especially impressed with its interpretation of material: silver reflects light believably, and gemstones refract along their facets. 

▫️Results: Translating 2D to 3D objects with Nano Banana and Midjourney is not only achievable, it’s pretty easy once you have a process. I don’t know what the 3D modeling side of Ai looks like, but as a non-modeler, this opens up incredible opportunities for small scale ideas from UI elements to vanity consumer shots. I can only presume that in the (near) future you could build viable 3D models from barely-there 2D assets. 

🔴If You’re Anti Ai: The tools can still be used to previs 3D models, or to inspire designs for what is “hidden” behind concept art. You can also successfully argue that although the results are impressive, they’re impossible to edit with any granularity without a real modeler. Additionally, these results can’t be saved pit as a 3D model… yet.

🔵If You’re Pro Ai: The possibilities are endless. From vanity shots of etsy wares to rotating logos to wireframe schematics. So long as the scope and complexity are appreciably small, you can add a wealth of polish to any of your flat 2D creative endeavors. Just be wary that the Machine can override your prompts with its Robocop-style Directives on perfect loops. 

All the source files can be found here so you can experiment yourself (right click save as).