Shutterstock launched a cloud-based generative 3D API, built on NVIDIA Edify AI architecture, trained on licensed Shutterstock content, as a fast way to produce realistic 3D models with AI.
3D Scene models
Stock media licensing website Shutterstock has launched and trained a generative 3D API, built on the NVIDIA Edify generative AI architecture. The cloud-based API serves as a fast way to produce realistic 3D models with AI, and was trained within a 16-month period on Shutterstock content, including more than half a million 3D models, more than 650 million images and detailed Shutterstock metadata.
The Edify architecture is multimodal, able generate content from text or image prompts. When building 3D scenes for exploratory and prototyping purposes, using the Shutterstock API service speeds up the time-consuming, if creative, process of 3D scene-building – creating background objects to create a detailed scene, finding or building an evocative background and selecting a lighting map.
Integrating a Generative Approach
Shutterstock has released this text-to-3D and image-to-3D generator API platform so that teams, studios and other organisations can use generative AI in a customised way to accelerate pipelines, populate virtual worlds or bring a generative approach to the 3D tools their artists use regularly.
Scene design with lighting
The API can be integrated directly into users’ applications or pipeline, aiming to reduce the time from concept to creation. The 3D generation process starts by presenting a 2D preview image to the user, who can approve or change it. Once approved, the generator outputs the 3D files, which are supplied in GLB, USDz and OBJ formats as triangles and quad meshes, complete with physically based rendering (PBR) materials.
With each successive training cycle over the past 16 months, Shutterstock notes that quality and prompt adherence has improved by a significant margin, leading to assets that are suitable to use for planning and prototyping, previs and as a route to final designs.
Shutterstock is also announcing the start of early access to its 360 HDRi generator API, also built on NVIDIA Edify, which generates detailed natural environments for lighting 3D scenes. These new services were shown live during the Enhancing 3D Pipelines with Generative AI industry session held at the SIGGRAPH 2024 conference held 28 July to 1 August.
Scene materials
Ethics
Dade Orgeron, Vice President of Innovation at Shutterstock stressed the ethical nature of the API service, which has been entirely trained on licensed data as the way to ensure fair compensation for the original creators who also have the option to opt-out. Due to the recent, worldwide expansion of 3D, concerns have emerged that companies under pressure will not be aware of how generative AI models have been trained.
As part of this process, Shutterstock is also putting its own TRUST safeguards in place, which includes ethical compliance, legal integrity and brand safety for customers and their audiences through customised protections and content controls. This also includes implementing guidelines for preventing inappropriate content and content that resembles major brand IP.
To further help users, the API is launching with documentation intended to help developers and non-developers integrate it in a short time, as well as lightweight plugins and code samples for Blender and Gradio.
Video game iso view
At SIGGRAPH 2024, some of Shutterstock's early beta testers such as HP, Trigger XR and WPP demonstrated how they use the generator in various use cases. HP showed 3D prints and physical prototypes, Trigger XR demonstrated asset creation for XR environments using an Apple Vision Pro, and WPP applied generative 3D to virtual production in the NVIDIA Omniverse platform. Studios, Shutterstock's in house production service, also presented their immersive production capabilities including some real life examples for commercial clients.
For more information on the API, see the generative 3D landing page. You can also sign up to be notified on launch and gain access to the developer platform to preview API documentation. www.shutterstock.com
