Top 7 3D Asset Generative AI

AI is appearing in different industries, and we are excited about the future of 3D asset generation. In this article, we introduce 7 AI tools that can help game developers and studios to create high-quality 3D assets at lower costs.

For game developers and studios, 3D assets are often one of the most challenging parts of the development process, and can easily become a bottleneck. The cost of producing a model ranges from $60 to $1500, and takes 2 to 10 weeks to create. The cost of high-fidelity 3D models is very expensive - both in terms of production cost and usage cost.

With the help of generative AI and 3D asset management platforms, both of these costs can be reduced. As AI can generate 3D assets at an amazing speed, the demand for storing these assets is increasing, and 3D asset management platforms can assist with asset warehousing and content delivery. Although new 3D AI tools emerge every day, here are some tools that we are currently focusing on:

1. NvidiaGet3D - This generative AI uses simple 2D images for training, and can generate 3D shapes with high-fidelity textures and powerful geometric details. They are generated in a universal format, so models can be exported and used immediately. Get3D can generate any type of 3D object, whether it's a building, vehicle or character. They take a stance on generative AI to fill in the lack of details in a 3D environment, thus lowering the credibility of a scene. They believe that with AI, small details that require a team's massive resources to create can be perfected in minutes. For example, randomly generating non-repeating vehicles or characters with believable behavior in a large crowd. Sanja Fidler, Vice President of AI Research at NVIDIA and the head of the Toronto AI lab that created the tool, said: "GET3D brings us one step closer to democratizing AI-driven 3D content creation. It can instantly generate textured 3D shapes, which could change the game for developers, helping them quickly populate all kinds of interesting objects in virtual worlds." Get3D is open source and can be found on Github.
2. 3DFY.ai - This text-to-3D generator can create high-quality 3D models. Assets are generated at multiple levels of detail (LOD) and have high-quality UV mapping to fit your project scope. Their process involves preprocessing, analyzing, and synthesizing. At the back-end, text or images are standardized and cleaned. For text, undefined tags are removed, and input is translated into more machine-readable language. Specifically, for images, they are cropped and isolated from the background by removing it. Then, the data is rendered into target code, which finally generates 3D assets. You can check out some of their 3D models here.
3. Sloyd.ai - Sloyd is designed for gaming. It is a fast online tool for automatic 3D asset creation. Every model is unwrapped and optimized for real-time use, so it can be directly integrated into your project. The Sloyd SDK can be used to create 3D assets in real-time in different environments. Users can generate procedural games or simulations based on their rule set. Sloyd has a generator library that can be customized for your specific project, allowing developers to flexibly choose when and how to generate 3D assets in real-time. Both on the server-side and user-side, the Sloyd engine can generate millions of vertices in less than 33 milliseconds. Each asset comes with detailed information that matches the number of vertices, and since images are generated rather than stored, they can help save storage space.
4. Gepetto.ai - Gepetto's philosophy is slightly different, as they are more interested in building new types of AI-centered games and movies rather than optimizing 3D asset production. Their flagship feature Collodi can "create a holographic deck-like experience for game players." They also created movie-making tools that can automate animation, deep fakes, dynamic observation, visual effects, and special effects. For gaming, they are developing AI game tools that are capable of infinite sandbox games, AI-driven role-playing games, complex dialogue mechanisms, and game worlds generated based on game decisions, which is very impressive. Generative AI reduces the amount of output teams need to achieve for each level by orders of magnitude.
5. LumaAI - LumaAI's Imagine3D tool allows you to input text to generate fully solid 3D models with full-color textures. LumaAI is said to produce higher quality 3D assets than some competitors, as it uses real-time imaging as a reference. What sets LumaAI apart is that it is designed for iOS devices, so users can generate 3D assets in the environment they are already familiar with in the real world. Users can generate 3D models with their own 2D images and can edit animations and other details in a web application. This provides a fast way for creative AR applications. Watch the video below to see how they create portals through real doors using hardware triggers.
6. MasterpieceStudio - MasterpieceStudio has 3 simple steps: generate, edit, and share. They claim to have built "the first generative AI" to create game-ready 3D assets. The goal is to reach 1 billion 3D creatives. That's a lot of 3D asset management! The MasterpieceStudio platform has a full set of tools to help creators generate usable 3D assets. This solves the problem of searching for random 3D resources in a library or resource pack, whether it's a conversion issue or a UV mapping issue. The resources made in MasterpieceStudio are ready to use.
7. Google DreamFusion - Google's 3D model generative AI - DreamFusion - does not require training on 3D model data, but the 3D models it generates are slightly different from other platforms and are not the preferred tool for game development. The system uses 2D images of objects generated by the Imagen text-to-image diffusion model to understand the different perspectives of the model it is trying to generate. This process is referred to by Google engineers as Score Distillation Sampling (SDS). SDS creates the basic appearance, and DreamFusion optimizes the asset to fill the model, such as adding regularizers and improving geometric shapes. After processing, these models have high-quality normals and can be lit up like regular 3D models.