For the last 60 days or so I've been working on a
@threejs based Ultima fan project in my spare time. I thought today I would share some of the workflow that helps generate consistent assets within
@MeshyAI .
Originally I started by having Opus 5 help me write an extraction tool for all of the Ultima 7 game data, which is what this project is grounded on. U7's game data is stored in a fairly logical way so it was easy to extract effectively every single monster, object, tile, etc out of the game. This wound up being significantly easier than I had imagined.
This data was then broken into "Bestiary", "Larder", "Objects", "Characters", etc to help organize and keep each system manageable. We then generated a prompt for each object using the game data as a starting point. Obviously Opus 5 hasn't played Ultima 7 so some of these prompts were adjusted, using my own imagination as well as references from D&D monster books, historical documents, etc.
I then use Meshy's Nano Banana based image generator to create a set of 2-4 images from each objects prompt. Multiple images allow me pick the best of, or to add some mild variation into the world. One of the keys to consistent generation I have found is to simply tell it you're using it for 3d modeling! "Soft even lighting, no cast shadows, plain flat neutral background, centered composition, suitable as reference for 3D modeling. One creature only. No base, plinth, rock, ground plane, scenery, water, fire, smoke, rider, saddle, weapon, text or human."
At that point we convert the image into a 3d model directly inside Meshy. Meshy will render a super complex model, which we then re-mesh down to a manageable 10k triangles and then texture. From here I'll export a GLB or FBX file, and either bring it into Mixamo for rigging or have Astra with a local Blender MCP animate it. I've found that giving Astra slow motion reference videos will generally result in usable output.
The final photo shows our Mongbat in the sewers beneath the city next to a giant sewer rat, since we're in the sewer the entire scene is a bit dark. Using this workflow allows me to compress what was once many days of work down into about 20-30 minutes, depending on how much touching up the model and animations need.