Omni123: Exploring 3D Native Foundation Models with Limited 3D Data by Unifying Text to 2D and 3D Generation
2026-04-02 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial Intelligence
AI summaryⓘ
The authors developed Omni123, a model that can create both 2D images and 3D objects from text in one system. They use a clever trick to connect images and 3D shapes by turning everything into tokens the model can understand together, allowing the model to learn 3D shapes better using lots of 2D image data. Their training mixes different tasks so the model learns to keep the meaning, look, and 3D structure consistent without needing perfect matching data. Tests show this approach helps create and edit 3D objects guided by text more accurately.
multimodal large language modelstext-to-image generationtext-to-3D generationautoregressive frameworkcross-modal consistencygeometric consistencydiscrete tokenssemantic alignmentmulti-view geometry3D synthesis
Authors
Chongjie Ye, Cheng Cao, Chuanyu Pan, Yiming Hao, Yihao Zhi, Yuanming Hu, Xiaoguang Han
Abstract
Recent multimodal large language models have achieved strong performance in unified text and image understanding and generation, yet extending such native capability to 3D remains challenging due to limited data. Compared to abundant 2D imagery, high-quality 3D assets are scarce, making 3D synthesis under-constrained. Existing methods often rely on indirect pipelines that edit in 2D and lift results into 3D via optimization, sacrificing geometric consistency. We present Omni123, a 3D-native foundation model that unifies text-to-2D and text-to-3D generation within a single autoregressive framework. Our key insight is that cross-modal consistency between images and 3D can serve as an implicit structural constraint. By representing text, images, and 3D as discrete tokens in a shared sequence space, the model leverages abundant 2D data as a geometric prior to improve 3D representations. We introduce an interleaved X-to-X training paradigm that coordinates diverse cross-modal tasks over heterogeneous paired datasets without requiring fully aligned text-image-3D triplets. By traversing semantic-visual-geometric cycles (e.g., text to image to 3D to image) within autoregressive sequences, the model jointly enforces semantic alignment, appearance fidelity, and multi-view geometric consistency. Experiments show that Omni123 significantly improves text-guided 3D generation and editing, demonstrating a scalable path toward multimodal 3D world models.