Subvolts take Meta's AssetGen aims to democratize 3D world creation from text prompts. Sep 19, 2025
For you
Developers
Learn about Meta's approach to democratizing 3D content creation with AssetGen and its architecture.
Business
Understand how AI advancements in 3D generation could impact virtual environments and content production.
Hype note
3/5Notable

This episode offers insight into Meta's research and development of a novel AI model for 3D content generation.

Meta Tech Podcast · Ep 78 Episode Official · Meta

78: Generating 3D Worlds with AI

36 min Interview Hosts: Pascal Harris
0:0036:11

Audio streams from Meta. Your place is remembered on this device.

From Subvolts

Summary and analysis by Subvolts editorial, AI-assisted.

Summary

Meta Tech Podcast episode 78 features Mahima and Rakesh discussing AssetGen, a foundation model for 3D content creation. They explore challenges in training with limited data and how large language models offer solutions. The conversation also covers generating entire 3D worlds from text prompts, aiming to democratize 3D asset creation.

Explained simply

Imagine creating entire 3D worlds just by typing a description. This episode explains how Meta is building a special AI model, AssetGen, to make this possible. It's like giving computers the tools to build virtual places from your imagination.

Why listen

This episode is for developers and creators interested in the future of 3D content. Mahima and Rakesh detail Meta's work on AssetGen, a model designed to simplify 3D asset and world generation. It connects to current frontier lab efforts in generative AI and the metaverse, showcasing Meta's push to make complex 3D creation more accessible.

On the mic
  • HostPascal Harris
  • GuestMahima
  • GuestRakesh
About
Key facts
  • Episode Title: 78: Generating 3D Worlds with AI
  • Show: Meta Tech Podcast
  • Length: 36 minutes
  • Model Focus: AssetGen, a foundation model for 3D
  • Key Technology: Large language models unlocked key solutions for training
Show notes from Meta

Creating 3D assets can be daunting, but does it have to be? Mahima and Rakesh are on a quest to democratize 3D content creation with AssetGen, a foundation model for 3D. They discuss the challenges of training such a model given the scarcity of available data and how large language models have unlocked key solutions. As if that weren't enough, they're also tackling the ambitious goal of generating entire worlds from a simple prompt. Tune in to learn more!

Got feedback? Send it to us on Threads (https://threads.net/@metatechpod), Instagram (https://instagram.com/metatechpod) and don't forget to follow our host Pascal (https://mastodon.social/@passy, https://threads.net/@passy_, @passy.bsky.social). Fancy working with us? Check out https://www.metacareers.com/.

Links

Timestamps

  • Intro 0:06

  • Introduction Mahima 1:39

  • Introduction Rakesh 2:57

  • Team mission 3:26

  • Why is 3D content hard to create? 5:15

  • The Metaverse 7:49

  • Tooling vision in Horizon Worlds 10:31

  • AssetGen Architecture 15:27

  • Consolidating models 18:25

  • From assets to worlds 19:22

  • Time to generate 24:46

  • Feedback loop 26:41

  • What's the market for AssetGen 29:49

  • What's available today? 31:26

  • What's next? 32:11

  • Outro 35:24

Jargon decoded
foundation model
Mahima and Rakesh
A large, versatile AI model trained on a broad range of data that can be adapted for many specific tasks.
Published
Show
Meta Tech Podcast · Meta
Length
36 min
Companies
Meta
Products
AssetGen, Horizon Worlds
Summary by
Subvolts, using an AI model (how we work). Spotted a mistake? Tell us.

Questions people ask

What is AssetGen?
AssetGen is a foundation model for 3D content creation that Meta is developing to simplify the process of generating 3D assets and entire worlds.
What challenges did Meta face in building AssetGen?
The scarcity of available data for training 3D models was a significant challenge, which large language models helped to overcome.

Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Meta.