ZTX
As Lead Technical Artist at ZTX, I set standards, supervised, and did hands-on automation work across every production stream to help bring Asia's largest metaverse, Zepeto, to web3. Collaborating with a talented team of artists, engineers and product managers, I ensured assets were successfully planned, executed, and integrated into the game and web3 ecosystem. Through projects like Placeables, Wearables, Genesis Homes, Arcadia Land, and marketing renders, my team delivered 10,000+ 3D assets.
Technical Direction and Asset Pipeline
The first order of business was to define standards for various asset types. I created training videos, documentation, templates, and folder structures for both the internal and outsourced art teams while considering future automation. This was followed by setting up centralized storage with auto-backup and asset versioning.
Genesis Homes
This workstream required converting 4000 procedurally generated NFT homes into game-ready assets. I worked closely with the Houdini team to develop ways to optimize meshes and extract simplified colliders, interaction points, and materials. Editor scripts and rigid naming conventions allowed me to manage and automate the entire prefab assembly and asset bundle creation. This NFT collection sold out on day one.
Art Pipeline Automation
To handle a large number of assets I wrote custom exporters for Max, Maya, and Substance Painter. Coupled with Unity Editor Scripts, I was able to create and manage thousands of prefabs and rebuild them in bulk when needed. Creating custom art tools was essential for maintaining a fast pace and consistent quality.
Thumbnails with Exact Margins
We had to create previews and thumbnails for thousands of furniture items directly from Unity. There was an issue - RenderTexture does not include post-processing and color correction, ScreenCapture does but it lacks alpha. I opted to capture using both methods and combine two sets of images using Python. Additionally, Python handled cropping images to the exact margin and output to multiple sizes and formats.
QA Automation with Computer Vision
Doing visual QA for 4,000 homes is no easy task, and after the first round, it became apparent that it's not 100% reliable and is very time-consuming. I created tools that leveraged OpenCV to detect Z-fighting and missing architectural elements. We captured and processed 96,000 images to create a list of failing assets. Custom Maya tool allowed artists to quickly go through the list of flagged assets and apply fixes.
