Announcing RNP-011: Piwa — Creating Millions of 3D Training Assets
Today, Emad Mostaque, Founder of Stability AI and Advisor to the Render Network, shared RNP-011: Piwa — Creating Millions of 3D Training Assets with the Render Network community for feedback.
About Piwa — Creating Millions of 3D Training Assets
The new RNP proposes a model for Stability AI to leverage the Render Network to create a data set of one-billion 3D assets used to train the next generation of 3D artificial intelligence.
According to the proposal, usage of the data set will be mediated exclusively with $RENDER, and Render Network Nodes will be incentivized to contribute to asset generation with a new staking program that helps administer governance and rewards nodes for providing upfront compute and storage infrastructure.
“Stability AI will work with the Render Network to create a dataset for 3D asset training. Leveraging the multi-actor composition of the Render Network, nodes, users, and community members will be able to participate in the creation (and benefits) of this dataset. To train models using this dataset, Buyers will need to license the dataset with $RENDER tokens.”
The proposal seeks to train the next generation of AI, built from the ground up on natively generated 3D data sets, creating the inputs for better models that are portable to emerging 3D applications in immersive media.
Language and image datasets have found success with scaling, but other forms of data have been left behind. 3D datasets like ShapeNet — the only comparable data source in the realm of 3D — are exclusively advanced by individuals manually creating and sharing assets. In short, the process for training 3D models is nearly impossible to scale. Data scarcity has become a bottleneck for learning-driven methods in 3D artificial intelligence. Previously Stability AI funded and supported the creation of Objaverse-XL https://stability.ai/research/objaverse-xl-a-colossal-universe-of-3d-objects, the largest publicly available 3D dataset which has enabled the latest wave of generative 3D models. High quality multiview 3D opens up improved image, video and virtual world models. These in turn can create better data that make the next generation of models even better, a powerful feedback loop to enable fully immersive media, interactive 3D and increased photorealism.
Models trained in Piwa will use the Render Network’s provenance system — Proof of Render and Proof of First Upload — hashing and recording 3D provenance metadata as compressed NFTs on Solana.
Provenance information for generated 3D data will be recorded as compressed NFTs on Solana. All assets created during the 3D data generation run on the node will be immediately hashed and sent to the Render Network ledger and file system (users can optionally package data as an ORBX geometry node asset) effectively generating a Proof of First Upload for assets to be used in future Render jobs or training jobs. This process will tie the generated assets to Render Network node operator’s wallet, and a Solana compressed NFT will be utilized to store and track 3D metadata.
Today, the Render Network generates a current Proof of First Upload for all texture, mesh and auxiliary scene datasets used for Render jobs. Proof of First Upload is just one layer of provenance leveraged for royalty and attribution systems such as the RNP proposed here, with the second being pre-hashing of assets that override POFU for known, high reputation 3D data sources (KitBash3D, LightStage, etc.).
After the successful generation of the dataset, Buyers interested in running models on the dataset can do so through an interface that allows them to pay for the data and license it for a given amount of time. Users will need to pay in RENDER, and revenue generated from will flow back to Nodes and Users who participated in the generation of the dataset.
Community Discussion for RNP-011 is Open
This RNP is open for comments and discussion from the community before being setup for a Community Vote.
In-depth details about implementation for Piwa, which includes the staking, governance, and economic model can be found here: https://github.com/rendernetwork/RNPs/blob/main/RNP-011.md
Discussion will take place on our Discord #RNP011 https://discord.gg/StZtjzCT
Join us in the Rendering Revolution at:
Website: https://render.x.io
Twitter: https://twitter.com/rendernetwork
Knowledge Base: https://know.rendernetwork.com/
Discord: https://discord.gg/rendernetwork
Telegram: https://t.me/rendernetwork
Render Network Foundation: https://renderfoundation.com/
