NC AI will unveil the next-generation model of its artificial intelligence (AI) service VARCO 3D, which generates 3D assets based on images and text.
NC AI said on the 24th that it has completed development of the next-generation model of its 3D generative AI service VARCO 3D, "VARCO 3D 2.0." VARCO 3D 2.0 is a model that improves the performance of the previous version based on user feedback accumulated since its launch in Dec. last year and the company's own research results.
The new model focuses on reducing distortion of the original shape, cited as the biggest limitation of 3D generative AI. Previous 3D generative AI frequently produced results that morphed into forms different from the source image or collapsed detailed structures, limiting real-world production use.
According to NC AI, VARCO 3D 2.0 can reproduce even complex decorative elements and detailed structures while reflecting the silhouette and overall proportions of the source image. It said objects with complex structures such as characters, creatures, and buildings can be precisely realized while maintaining the original design intent.
Benchmark performance has also improved. It scored 0.449 on Uni3D, a representative shape-similarity metric, an improvement of about 40.8% over version 1.1 (0.319), and it outperformed major competing models, including the latest global open-source model Trellis2 (0.436), Ultrashape (0.428), and Hunyuan3D 2.1 (0.427). It also recorded the highest performance on key metrics such as CLIP-N and ULIP-2.
Visual completeness has also been upgraded. It supports high-quality textures up to 4K resolution and realistically renders everything from the color and material of the source image to the textures of various materials such as metal and wood, as well as fine surface patterns and wear. This provides results ready for immediate use across fields that require high-quality 3D modeling, including beyond games and video content to industrial simulations.
NC AI said VARCO 3D can also serve as a foundational technology for physical AI and Digital Twin implementation. For physical AI to understand and learn the real world, it needs large-scale 3D data and virtual environments that reflect the shape and physical properties of real objects.
VARCO 3D provides a foundation to quickly build digital objects that can be combined with physical property information. According to NC AI, 3D assets that previously took experts four weeks to produce can be generated in as little as three minutes, reducing the time and expense required to build datasets.
The VARCO 3D 2.0 model will be applied to the VARCO 3D SaaS service in July. Various new features will be updated together, and existing users will be able to experience the VARCO 3D 2.0 model in the same service environment without any additional procedures.
Lee Yeon-su, head of NC AI, said, "VARCO 3D 2.0 will transform content creators' workflows with unrivaled shape fidelity and high-quality texture expression," and added, "We will not only secure data security and AI technology independence based on our proprietary AI, but also grow it into core infrastructure leading the era of physical AI and Digital Twin."