Modelyth builds AI-native 3D models that generate clean, commercial-safe assets from text or images. We leverage recent research (e.g., TRELLIS-style multiview pretraining) to deliver meshes and textures you can ship—across games, film/VFX, and design.
Text prompt or reference images
Watertight GLB/FBX/USD with PBR textures
Multi-view coherence, low fix-up time
Games • Film & TV • Previz • eCommerce • XR
Curated, license-clear datasets with audit trails.
Clean import into UE5/Unity, Maya/Blender/Houdini.
Architecture designed for large 3D corpora.
Modelyth accelerates world building. Artists and developers describe what they need; our model returns clean, production-ready 3D assets that pass technical checks (scale, normals, topology). Our mission is to reduce asset creation time from days to minutes while maintaining commercial safety—whether the target is a game level, a TV episode, or a feature film.
Trellis-style multiview pretraining with structured latents at 512², optimized with BF16 + fused attention.
Watertight meshes, packed UVs, and PBR textures (2K–8K). Export to GLB/FBX/USD; optional auto-retopo and LODs.
Curated training data (CC0/commercial-safe), versioned preprocessing, and per-asset provenance for enterprise audits.
Production inquiries, pilots, or dataset/license questions: