Artificial Intelligence in 3D Modeling
From digital sculpting, to simply taking a video of an object, there are a variety of ways to create a 3D model. A 3D model is created in a Computer Aided Design (CAD) program that is made up out of connected points in space. There are a variety of ways to construct a model, but in the end the model is composed of either triangles or Non-Uniform Rational B-Splines Modeling (NURBS). The main difference is that a model constructed out of triangles is composed of millions of small flat surfaces, while a NURBS model allows for mathematically perfect curves (What Is NURBS, n.d.). While 3D modeling has been around for a long time, it has become much more popular in the last 20 years since the release of the first fully Computer Generated Image (CGI) film, Toy Story. The first three-dimensional design program, Sketchpad by Ivan Sutherland, was a primitive program that laid the basis for a lot of future 3D modeling (Pyfer, 2017). This has caused the development of a slew of modeling programs, from ones that specialize in parameters like Fusion 360 to ones that are more freehand like Blender. Another difference between these two programs is that Fusion uses mainly NURBS, while Blender manipulates triangles. This speaks to their two uses with Fusion needing mathematically perfect models for real world construction, and therefore uses parametrically based NURBS, while blender does not need that precision and can operate with triangles (What Is NURBS, n.d.)..
Other 3D softwares have used a technique called photogrammetry to create models. For photogrammetry all a user has to do is take out a phone and take 150 pictures of an object at different angles, wait 30 minutes, and then they have a to-scale 3D model of whatever they want with a texture developed from the pictures.The way this works is the camera records its position relative to the starting position. Then, using that data, the algorithm draws virtual rays to the object and then constructs the model based off of the intersections of the rays ("Photogrammetry: Step-by-Step," n.d.).
For example, below is a chair that was created with roughly 120 pictures from various angles. One challenge for photogrammetry is that for larger objects it is hard to accurately take pictures of the bottom and make a solid object. Additionally it can be hard to ensure that enough different angles are covered. Not getting enough viewpoints can lead to models with inaccurate geometry, such as the model of a face below. While it is convenient to just need a couple of photos and 30 minutes of wait time to get a 3D model, it is not a good method for creating original models and designs.
One way to quickly create an “original” model is AI 3D modeling programs. There are many different ways to access AI in 3D modeling, either online or through an app. On the internet there are two main types of AI 3D model creators: Large Language Models, like ChatGPT, and dedicated AI model creators such as Meshy AI. Using a dedicated AI model creator is more advantageous than using ChatGPT because of the way the programs are coded. ChatGPT is a large language model that has a code compiler, allowing it to write either in ASCII or Python, which can both be used for 3D modeling. A Standard Tessellation/Triangle Language (STL) file, a common 3d model file type, is often written in ASCII and is a standalone file. Because all a STL file is is a list of points in a 3d space, the data can also be stored in python. Unfortunately 3D modeling software can’t read python as a 3D file, so a model made using Python has to be put into another software that can then run the code and export it as a standalone file. Without a paid ChatGPT subscription, there is a limited number of standalone STL files that can be created, so ChatGPT more often than not will write a Python code which can be put into Blender.(Utley 2023)
One of the biggest jumps forward for AI in 3D modeling is being able to analyze 3D data and generate a mesh based on that. Meshes are 3D models made out of nodes. Nodes are points that have an x,y, and z coordinate attached to them. Having quads instead of irregular shapes matters to 3D modelers who are planning to animate their models, especially faces, because it allows the mesh to deform in a predictable and natural way (Quad-Based Topology, n.d.). While ChatGPT can walk someone through remeshing or creating a model, asking it to create one itself creates a much worse-looking model as shown in the images below.
AI 3D modeling isn’t just limited to the more freehand text or single image generation, though. One 3D design program, Fusion 360 has had a parameter-based AI model generator since 2018. Unlike the other types of 3D model generators, Fusion 360 is designed so that the designers are able to tweak everything the AI does with sliders, and the generator model is designed to optimize cost and material usage rather than create an image that fits a prompt (Rathbone, 2018). There are two ways to access Fusion 360's generative design model, and they both function slightly differently. One way to access it is through the main design tab, where there is a button labelled “generative design.” If clicked, a prompt shows up that asks the user to click on faces that they want to connect together and which to avoid. Once all the faces have been selected, the user can hit “generate.” This generates six different ideas with a slider that can control the size and thickness of each design. Once the user has tweaked the design to look like what they want to, they click “done” and the generative design model outputs the finished 3D model. The other, more complicated and more customizable way to access the AI modeling feature is by going to the generative design workspace. To create a 3d model the user has to specify where force will be applied to the model and put in parameters such as which faces can be edited and what bodies should be avoided.
Another area where AI has advanced is in remeshing programs. Remeshing is where a 3D artist, usually for a game or movie, goes through the nodes of the object they created and cleans up the topology. This means going through and deleting or adding nodes so that the object is only made out of quads. One AI tool that has greatly reduced this is Nvidia’s Meshtron. Meshtron uses AI to analyze a mesh and then remesh it into having clean topology, or node setup. (Hao et al., n.d.). In addition to Meshtron, Nvidia has created Omniverse, a picture-to-model AI (Viviani, 2023). Both of these tools help make the process of turning ideas into design much quicker and less tedious for creators.
Advancements in 3D modeling AI have caused fears that 3D artists are becoming obsolete. This worry is unfounded because, while AI 3D tools can speed up the process and help with many steps, they are not powerful enough to eliminate humans entirely (Viviani 2023). Additionally, humans are required to specify the prompts and edit the models afterwards. Overall, while AI 3D modeling tools can speed up the 3D workflow, they are still merely a tool to be employed along with other programs and are not powerful enough to stand on their own yet.
References
Hao, Z., Romero, D., & Liu, M.-Y. (n.d.). High-Fidelity 3D Mesh Generation at Scale with Meshtron. Developer.nvidia.com. https://developer.nvidia.com/blog/high-fidelity-3d-mesh-generation-at-scale-with-meshtron/
Photogrammetry: Step-by-Step Guide and Software Comparison. (n.d.). https://formlabs.com/blog/photogrammetry-guide-and-software-comparison/
Pyfer, J. (2017, June 28). Sketchpad. Encyclopedia Britannica. https://www.britannica.com/technology/Sketchpad
Quad-Based Topology. (n.d.). Turbosquid. Retrieved March 30, 2025, from https://resources.turbosquid.com/training/modeling/quad-based-topology/
Rathbone, E. (2018, May 8). It's here! Generative design technology makes its commercial debut in Fusion 360 Ultimate. Blogs.autodesk. https://blogs.autodesk.com/advanced-manufacturing/2018/05/08/generative-design-technology-makes-commercial-debut-fusion-360-ultimate/\
Utley, E. (2023, March). A Chat with an AI CAD Designer via ChatGPT. Tech Briefs. https://www.techbriefs.com/component/content/article/47772-a-chat-with-an-ai-cad-designer-via-chatgpt#:~:text=ChatGPT%20can%20make%20a%203D,but%20doesn't%20quite%20succeed.
Viviani, M. (2023, August 31). How AI-Generated 3D Models are Transforming 3D Pipelines. Medium.com. Retrieved May 19, 2025, from https://medium.com/%40nvidiaomniverse/how-ai-generated-3d-models-are-transforming-3d-pipelines-d4245566a289
What is NURBS modeling and why is it crucial to CAD software? (n.d.). Dassault Systems. Retrieved May 30, 2025, from https://www.3ds.com/store/cad/NURBS-modeling