Korean startup Recon Labs has launched 3Dpresso, a service that uses AI to create 3D models from videos of real-world objects. Users can download these models in OBJ or GLB formats for use in DCC applications and game engines.
Combining Photogrammetry with Advanced AI Techniques
Many tools exist for making 3D models from photos. 3Dpresso uses a mix of photogrammetry and advanced tech like Neural Radiance Fields (NeRF). The MetaRECON engine provides more precise reconstructions, especially for texture-less or reflective surfaces.
A Simple and Fast Process for Creating 3D Models
To create a 3D model with 3Dpresso, users upload a video of a 3D object recorded with a smartphone or digital camera. The video should show two 360-degree rotations: one from the front and one from the top. Higher resolution videos produce better results, and Recon Labs recommends using 4K quality.
Once uploaded, Recon Labs’ servers process the video, which takes less than 30 minutes. After processing, 3Dpresso emails a download link for the 3D model in GLB or OBJ format with 1024×1024 pixel textures. The model quality suits real-time applications, including games and augmented reality projects, though some manual cleanup might be needed.
Additional Features and Future Plans
3Dpresso offers AI-based texturing, allowing users to guide the visual style of textures with simple text prompts like “blue crystal” or “hyper-realistic.”
Recon Labs plans to launch a mobile app for easy video uploads directly from phones.
Visit the 3Dpresso YouTube channel for examples of suitable videos and final results. With 3Dpresso, creating 3D models of real-world objects is easier and more accurate than ever.