Few-step 3d generation speeds geometry creation with better texture accuracy
Channel Match: Calibrating Shape Latents for Few-Step 3D Generation
Computer Vision and Pattern Recognition
Summary
Creating detailed 3D models usually takes many slow steps, making it hard to generate high-resolution shapes quickly. The authors present a way to speed up this process to just a few steps, keeping the shape quality high. They also found a new problem where faster methods cause texture errors even though the shapes are still correct, and they fix it by adjusting the internal features that control textures without retraining the model. Their improvements make 3D generation both faster and better for realistic looks.
What this means in practice
- •For 3d asset creators: Create high-resolution 3D shapes faster while preserving quality for use in games and simulations.
- •For visual effects artists: Improve rendering quality of 3D assets by reducing texture artifacts caused during fast generation.
Authors
Di Yang, Zhihao Li, Yanhai Xiong, Yufei Wang
Abstract
Recent native 3D generators achieve high-quality asset creation by modeling structured 3D latents with flow-based transformers, but their high-resolution generation remains computationally expensive due to costly iterative sampling. In this work, we introduce a few-step distillation framework for native 3D generation, enabled by continuous self-consistency training with a custom JVP FlashAttention kernel and an analytic velocity regression objective for sparse structured 3D latents. To stabilize joint optimization, we further introduce an adaptive loss weighting strategy that automatically balances them without manual tuning. Beyond acceleration, we reveal a previously overlooked failure mode of few-step native 3D generation: sampling-induced latent distribution drift preserves geometric correctness but disrupts downstream PBR texture synthesis, causing texture artifacts on geometrically valid assets. To address this issue, we propose Channel Match, a training-free adaptive per-channel latent calibration method that restores the latent distribution without modifying either generator. We validate our approach on TRELLIS.2 at the $1024^3$ resolution setting. Our two-step model achieves $4.5\times$ faster geometry sampling while maintaining geometry quality comparable to the many-step teacher, and Channel Match further improves rendering quality by recovering the latent interface between geometry and appearance generation.