Researchers from Kuaishou Technology and Nanyang Technological University have developed MetaView, a diffusion-based framework for monocular novel view synthesis. The model integrates implicit geometry priors and metric depth to enable precise camera control and geometry-consistent 3D perception from a single image.
MetaView addresses the persistent 'scale drift' and spatial distortion issues that plague monocular AI video synthesis, providing the precise camera controllability required for professional B2B virtual production. By bypassing the need for explicit, computationally heavy 3D reconstruction pipelines, it offers a more flexible way to generate interactive 3D content from existing 2D video assets. This development is critical for platforms seeking to automate the conversion of static imagery into navigable 3D environments. Industry watchers should monitor whether these spatial consistency gains are integrated into Kuaishou's commercial Kling 3.0 platform to support professional cinematography workflows.
The introduction of MetaView coincides with a massive commercial scaling of Kuaishou’s AI efforts. In July 2026, Kuaishou reportedly prepared a 20.45 billion yuan ($3 billion) financing round for its Kling AI unit to facilitate a separate IPO, valuing the subsidiary at approximately $18 billion per TechNode and South China Morning Post. This restructuring follows a period of rapid growth; Kuaishou’s Q1 2026 earnings report cited more than RMB 650 million in revenue for Kling, a 300% year-over-year increase, including its participation in high-end productions like the television series House of David.
The underlying models used in MetaView reflect significant state-of-the-art shifts in 2025 and 2026. MetaView utilizes DepthAnything 3, which ByteDance-Seed launched in November 2025 as a unified transformer for spatial reconstruction. This family of models replaced disparate depth and camera pose architectures with a single forward-pass system, significantly reducing the overhead for geometric perception. Similarly, the 20-billion-parameter Qwen-Image-Edit backbone was a cornerstone of Alibaba’s open-source strategy in late 2025, specializing in dual-path semantic and appearance control before its features were merged into the broader Qwen 2.0 series in early 2026.
Broader industry trends at CVPR 2026 highlight a pivot toward real-time novel view synthesis and 3D Gaussian Splatting. Per recent reports from June 2026, papers like UniSHARP and SHARP from Apple and Insta360 have introduced feed-forward methods to synthesize photorealistic 3D representations from single photographs in less than a second. Together with Kuaishou's MetaView, these developments suggest a transition away from slow, iterative optimization toward instant, geometry-aware synthetic video generation suitable for real-time streaming and XR applications.
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source