Meta, formerly Facebook, has launched a new research paper on Meta 3D Gen AI (3DGen). It seems like a new breakthrough in generative AIs where users may be able to create realistic-looking 3D characters, objects, and models using nothing but text prompts. Meta blog shared more details about the new 3D Gen AI tool.
Meta claims that its text-to-3D asset generation tool, aptly called 3D Gen, can transform textual descriptions into detailed 3D models quickly, producing high-quality shapes and textures in less than a minute.
So far, image-generative tools like MidJourney can only generate 2D images from text prompts. It is primarily used for digital art, design, marketing, and more. However, Meta’s 3D has wider applications. It is ideal for industries requiring rapid and accurate 3D asset creation, such as video game development, virtual realities (VR), AR, and digital art.
Looking at the research paper and the videos that Meta’s AI team has shared, it seems that 3D Gen model can create short videos with the 3D asset in action. And all this in under one minute.
Without getting into the technical aspects of it, you will find that in Meta’s research paper, let’s see what it is and how it differs from existing models:
Key Features:
- High prompt fidelity, ensuring the 3D models closely match the text descriptions.
- Supports physically-based rendering (PBR) for realistic lighting effects.
- Generative retexturing allows customization of existing 3D models.
- Represents objects in view space, volumetric space, and UV (texture) space.
So how does Meta’s 3D Gen AI model stack up against existing 2D image generative models like Midjourney, for example?
Key Differences:
- Output Type:
- Meta 3D Gen: Creates 3D models.
- MidJourney: Generates 2D images.
- Complexity and Application:
- Meta 3D Gen: Used for complex applications requiring 3D models, such as VR/AR and gaming.
- MidJourney: Focused on generating images for digital art and design.
- Customization:
- Meta 3D Gen: Allows retexturing of 3D models using additional text inputs.
- MidJourney: Allows iterative refinement of images through textual adjustments.
- Rendering:
- Meta 3D Gen: Supports physically-based rendering for realistic lighting.
- MidJourney: Primarily focuses on artistic rendering styles.
- Representation:
- Meta 3D Gen: Represents objects in multiple spaces (view, volumetric, UV).
- MidJourney: Represents objects as 2D images only.
As you will notice, both text-to-image and text-to-3D assets generative AI tools have different markets and applications. There will be some overlap but in the end, should compliment each other. We will keep you updated when the tool is released for public beta.
