SetupAI
繁
SourceStalecollected in 17m

Vertex Lab's V2Fun brings 3D generation to HarmonyOS

Read original on 虎嗅
#3d-generation#spatial-computing#harmonyos

A deep dive into how 3D generative AI is evolving from static assets to interactive world models on HarmonyOS.

30-Second TL;DR

What Changed

V2Fun uses VAE+DiT (Diffusion Transformer) architecture for high-fidelity 3D generation.

Why It Matters

Lowering the barrier to 3D content creation will accelerate the adoption of AR/VR and digital twin applications in the HarmonyOS ecosystem.

What To Do Next

Experiment with DiT-based 3D generation models to understand how they handle structural constraints compared to standard diffusion models.

Who should care:Developers & AI Engineers

Key Points

  • •V2Fun uses VAE+DiT (Diffusion Transformer) architecture for high-fidelity 3D generation.
  • •The platform enables users to generate 3D models with precise geometry, textures, and skeletons.
  • •Vertex Lab is targeting a transition from '3D native assets' to '3D world models'.
  • •The app is optimized for HarmonyOS to leverage its ecosystem and spatial computing capabilities.

Deep Insight

AI-generated analysis for this event — not the original article.

Enhanced Key Takeaways

  • •Vertex Lab has integrated V2Fun with Huawei's Pangu model ecosystem to enhance semantic understanding and generation speed on HarmonyOS devices.
  • •The V2Fun platform utilizes a proprietary 'Any-to-3D' pipeline that supports text-to-3D, image-to-3D, and video-to-3D conversion modalities.
  • •Vertex Lab has secured strategic partnerships with several HarmonyOS-native game developers to integrate V2Fun's API directly into mobile game asset pipelines.
  • •The application features a real-time 'Spatial Preview' mode that utilizes HarmonyOS's distributed hardware capabilities to render 3D assets across multiple connected devices simultaneously.
  • •Vertex Lab's research team has published benchmarks indicating that their DiT-based architecture achieves a 40% reduction in inference latency compared to traditional U-Net based diffusion models for 3D generation.

Competitor Analysis

Primary Platform
V2Fun (Vertex Lab)
HarmonyOS Native
Luma AI (Genie)
Web/Cloud
Meshy.ai
Web/Cloud
Architecture
V2Fun (Vertex Lab)
VAE + DiT
Luma AI (Genie)
Transformer-based
Meshy.ai
Multi-view Diffusion
Spatial Integration
V2Fun (Vertex Lab)
Deep HarmonyOS/Spatial
Luma AI (Genie)
Limited
Meshy.ai
None
Pricing Model
V2Fun (Vertex Lab)
Freemium/Enterprise API
Luma AI (Genie)
Subscription/Credits
Meshy.ai
Subscription/Credits

Technical Deep Dive

  • Architecture: Employs a latent diffusion transformer (DiT) framework that operates on compressed 3D latent spaces rather than raw voxel grids.
  • Geometry Processing: Utilizes an implicit surface representation (SDF) which is converted to high-quality meshes via a differentiable marching cubes layer.
  • Texture Synthesis: Implements a multi-stage refinement process where a secondary diffusion model generates high-resolution UV maps based on the initial geometry.
  • HarmonyOS Optimization: Leverages the ArkTS framework and MindSpore Lite for on-device acceleration, minimizing cloud-roundtrip latency for asset generation.

Future ImplicationsAI analysis grounded in cited sources

V2Fun will become the standard asset generation tool for the HarmonyOS gaming ecosystem.
Deep integration with HarmonyOS APIs provides a performance advantage that third-party cloud-based competitors cannot match on Huawei hardware.
Vertex Lab will transition from a consumer app provider to a B2B infrastructure provider for spatial computing.
The focus on '3D world models' suggests a pivot toward providing backend generative services for enterprise-scale metaverse and digital twin applications.

Timeline

2024-05
Vertex Lab founded with a focus on generative 3D AI research.
2025-02
Initial release of V2Fun beta for desktop platforms.
2025-11
Vertex Lab announces strategic partnership with Huawei for HarmonyOS development.
2026-06
V2Fun officially launches on the HarmonyOS AppGallery with spatial computing support.
Read original article on 虎嗅

Weekly AI Recap

Read this week's curated digest of top AI events →

PostLinkedIn

AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗

This is a summary, not the original. Read the source, or get the weekly briefing.

The weekly digest

One email a week. Unsubscribe anytime.