Gen AI & ML
Embedded Files
AI-Generated 3D: Reality is Closer Than You Think! (2025)
AI-Generated 3D: Reality is Closer Than You Think! (2025)
The eCommerce industry increasingly relies on 3D content for digital twins and spatial experiences, but traditional 3D asset creation remains labor-intensive, requiring custom models derived from manual modeling, 3D CAD optimization, photogrammetry, or LiDAR scans. However, advancements in NeRF, Gaussian Splatting, and Generative AI are opening new possibilities. In particular, generative AI is making significant strides in 3D content creation, helping to overcome key production bottlenecks.
In our recent research, we explored the evolving landscape of generative 3D technologies, assessing AI models' capabilities, limitations, and potential workflow integrations. Here are our initial test results, featuring real-world product samples—check them out! You can also view the models in augmented reality on Android or iOS devices.
Here are our initial test results by sampling some real-world products - check them out here. You can also view the models in augmented reality using Android or iOS devices:
Insights
Insights
The displayed 3D models (GLB/USDZ) are direct outputs from AI prediction inference, with minimal adjustments to dimensions and colors. Geometry and texture are AI-generated. The initial tests on Home category objects (cabinets, chairs, sectionals, bench) provided valuable insights:
Few Images, Varied Angles: High-quality images with neutral lighting and diverse angles are crucial to produce good results, even when only a limited number (1-5) are available.
Clean Backgrounds: Effective background removal significantly improves inferences.
API Integration: The availability of API support streamlines integration into existing workflows.
Multiple Image and 3D Shape Support: Leveraging multiple images, and even 3D shapes in some cases, enhances model accuracy.
Feasible Cost and Time: Cloud and local compute inference times (3-6 minutes) make the technology increasingly practical for adoption.
Textured Output with Limitations: While all solutions generate albedo textures, automatic PBR and normal map generation still require improvement for added realism.
Takeaway
Takeaway
The rapid growth of eCommerce has fueled a demand for 3D content to create rich, engaging digital experiences. Recent advancements in generative AI for 3D content creation show great potential for increasing efficiency and quality. While the current output may not be suitable for all production needs, it can significantly reduce artist workload by providing a base geometry shape, serving as background stand-ins, or enabling rapid concept visualization. The future of 3D content creation is evolving at a fast pace. What are your thoughts on generative 3D? Let's discuss in the comments!
#3D #AI #GenerativeAI #Innovation #eCommerce #AR #3DModeling
Report abuse