We crammed a 24GB AI 3D-generation pipeline into a completely offline desktop app (and the Demo is live)

If you are an indie game developer right now, you know the pain of 3D asset generation. The current landscape of AI 3D tools is a nightmare of expensive monthly SaaS subscriptions, API paywalls, and cloud-based platforms that claim ownership over the meshes you generate. We got tired of it. At Odyssey Game Studios, we decided to build the anti-SaaS solution: Jupetar. It’s a completely offline, local-compute 2D-to-3D asset generation pipeline. You pay once, you own the software, and it runs entirely on your local GPU. After weeks of engineering (and battling the Steam review queue), the official Jupetar Demo is now live on Steam. Here is a look under the hood at how we built it, and how you can test it right now. The "No-Ping" Architecture The biggest challenge in building local AI wrappers is preventing the underlying scripts from constantly trying to phone home. Jupetar relies on a massive 24GB local models folder containing the Hunyuan DiT (geometry) and Paint VAE (textures) weights.