#001 - Pipeline B: how HyperKourier's assets get made

Today I locked in how HyperKourier's 3D assets get made for the next six weeks.

In February I sketched out an asset pipeline using Meshy for hero characters and Sloyd for modular buildings. Both real tools, both fine. The plan was: one tool for one job, lock in early, ship.

Three things changed between then and now.

fal.ai consolidated several frontier 3D models — Trellis 2, Hunyuan3D v2, Pixel3D, Tripo 3.0, Rodin Gen-2 — behind one MCP endpoint and pay-per-generation billing. You don't have to subscribe to each vendor to compare them. Higgsfield Plus covers both concept imagery and short-form video at one annual cost. And on May 12, TencentARC released Pixel3D. Too new to bet a pipeline on, but cheap to test alongside Meshy.

So the plan changed. I'm calling it Pipeline B because v1 was Pipeline A and I'm not clever enough to think of better names.

Pipeline B:

• Concept art and short video: Higgsfield Plus, annual, ~$39/mo
• 3D generation: fal.ai, pay-per-gen, ~$10-15/mo expected
• Cleanup and export: Blender 4.5 LTS, local
• Music: Suno, paid before launch
• SFX and barks: ElevenLabs, paid before launch

Total: ~$47-55/mo while the game is in active production.

What I'm going to do with $23 of fal.ai credits: run a head-to-head between Pixel3D and Meshy on five concept pieces. A KottiExpress courier. A CivicLoop officer. An architectural panel from Knoten Null. A vending machine. A neon sign. Score each on silhouette fidelity, cel-shade compatibility, UV unwrap quality, topology, and end-to-end time. Decision date: July 1.

There's one open question I'm punting on. Hunyuan3D 2.1, one of the cheaper fal.ai models, ships with a territorial license clause. My reading is that the clause targets self-hosting the weights, not consuming through fal.ai's hosted endpoint. I'm 80% confident. Not a legal opinion. Before any Hunyuan3D asset ships to the public preview, I'll get a real review.

Pattern I'm trying for the devlog: I post when I ship. Today I shipped a decision and a test plan. That's enough for one post.