Where imagination meets reality.
Cosmic Gen v1 — Private previewCosmic Gen by AnantrixLabs turns a single reference into rigged, textured, engine-ready 3D — production-grade geometry, PBR materials, 4K detail. Built for studios that ship.
Plugs into the tools you already ship with
How it works
Cosmic Gen runs a six-stage pipeline that turns a reference image into a textured, rigged 3D asset — without the manual cleanup that legacy generators leave behind.
Your reference image is analyzed for identity, structure, and intent — what you want, not just what's visible.
Geometry, materials, UV, and rigging are composed together in a single pass — every stage informed by the others.
Export to glTF, FBX, USDZ, or VRM. Drop into Unity, Unreal, Blender, or the open web.
The platform
From the reference image you upload to the rigged, textured asset you drop into your engine — all in one run. No tools to chain. No manual cleanup. No part of the pipeline left for someone else.
Watertight, manifold geometry with clean topology. Sharp features stay sharp; organic surfaces stay smooth. Built for studios that need to ship — not for demo reels.
Automatic UV layouts with intelligent seam placement. Editable, predictable, and ready for hand-painted detail if you want it.
Full PBR stack — albedo, roughness, metallic, normal — generated together. Drops straight into Unreal, Unity, or any glTF-compatible engine.
Auto-skinned to a standard humanoid or custom skeleton with 52 facial blendshapes. Retargets cleanly to your existing animation libraries.
Patterns, colors, faces, accessories — preserved exactly. No drift across views. No generic-looking outputs.
Crisp 4K maps out of the box. Tile-based super-resolution keeps fine detail sharp, edge to edge.
Use cases
Unity- and Unreal-ready FBX with bone hierarchy, blendshapes, and quad-dominant topology. Cut character pipeline time from weeks to hours.
Convert flat catalog photos into rotatable 3D product views. Photoreal PBR materials, deployed via glTF on the open web.
Optimized LODs and PBR pipelines for Apple Vision Pro, Quest, and the spatial web. USDZ and glTF export, baked or live shading.
Universal skeleton retargets to existing animation libraries. Spring bones and cloth sim built-in for production-grade motion.
Infrastructure
Cosmic Gen is trained on distributed GPU clusters and deployed across a multi-region inference fleet. We optimize aggressively for A10- and L4-class deployment to keep generation cheap at the edge.
Cloud-native serving
Inference runs on auto-scaling GPU pools across regions. Cost-efficient defaults, high-fidelity tiers on demand.
Always learning
Trained continuously on a growing 3D corpus and refined with feedback from studio partners.
Data pipeline
Streaming-first ingestion. License-clean, provenance-tracked, and built for scale.
Built for builders
Queue-backed jobs with async delivery. SDKs in Python and TypeScript. Predictable latency.
AnantrixLabs runs production on AWS — auto-scaling GPU compute, object storage for training corpora, and managed services for the model lifecycle. Architected cloud-agnostic so workloads can burst across providers as we scale.
Roadmap
End-to-end image-to-3D in private preview. Onboarding partner studios for the v1 release.
Native animation generation, simulation, and live retargeting onto generated rigs.
Fine-grained style control, multi-object scene composition, and tighter alignment with studio judgment.
Team
AnantrixLabs is a small team of ML researchers and graphics engineers — alumni of leading AI labs and 3D platforms — focused on closing the gap between generative AI and what ships in real games and products.
Generative 3D, diffusion, and large-scale training. Published at SIGGRAPH and NeurIPS.
Real-time rendering and DCC pipeline veterans — Unity, Unreal, Maya, Houdini.
Distributed training, GPU performance, and cloud-native deployment.
Early access
Tell us what you're building. We're onboarding a small group of studios, e-commerce teams, and platform partners this cycle.
Or reach us directly at hello@anantrixlabs.com