As interest in generative artificial intelligence (AI) continues to rise, the 3D printing industry has witnessed growth in text-to-CAD platforms. These tools allow users to create 3D printable CAD files from simple text prompts, making additive manufacturing more widely accessible.
Over the past year, Shenzhen-based 3D printer manufacturer Bambu Lab and Markforged’s founder Greg Mark have separately announced new AI 3D model generation software. Other platforms, like Meshy.ai, OpenAI Point-E, neThing.xyz, and Google’s DreamFusion, offer similar capabilities. Leading chip designer NVIDIA joined the industry-wide movement at the start of 2023 with its own generative AI tool.
Another company active in this space is Zoo. The Los Angeles-based software developer offers a CAD modelling app, model comparison extension for GitHub, and an AI-powered text-to-CAD tool. The latter, an open-source platform that generates CAD files from short text prompts, leverages the firm’s KittyCAD design API and ML-ephant machine learning API. These create 3D printable models by programmatically analyzing training data.
KittyCAD was the company’s original name when it launched in 2021. However, in January 2024, the Californian tech firm rebranded to Zoo. According to Co-founder and Executive Chairman Jordan Noone, this move came amid a split within the customer base. A growing number of developers were leveraging the design API to build proprietary tools, while other users were only interested in the Modeling App and Text-to-CAD capabilities. “This has left us with a challenge – these two customer cohorts are looking for very different things,” he stated in a blog post.
Noone, the Co-founder and former CTO of aerospace manufacturer Relativity Space, explained that Zoo is now the company’s umbrella brand. KittyCAD survives as the name of its flagship design API.
KittyCAD becomes Zoo
KittyCAD launched in 2021 to develop software and API for hardware design. The company set out to build the “CAD of the future: fully programmable, user friendly, multi-threaded, and fast.” Backed by Embedded Ventures, Noone’s deep tech venture capital firm, the startup has claimed to be “the world’s first and only company to build hardware design infrastructure for the internet.”
In addition to a pre-seed financing round led by Embedded, KittyCAD was also part of NVIDIA’s Inception program. This initiative is designed to help startups achieve growth and innovation during product development, prototyping, and deployment. It offers access to NVIDIA expertise, while facilitating connections with venture capitalists, AI organizations, and industry experts.
During KittyCAD’s 2024 rebranding and the introduction of Zoo, the new-look company also unveiled its machine learning API, ML-ephant. The interface reportedly bridges the gap between machine learning and hardware design by “shortcutting from design intent to a CAD model.” It is trained on the company’s proprietary data and uses the KittyCAD Design API. This is built on Zoo’s GPU-native Geometry Engine, allowing the machine learning API to generate CAD files by programmatically analyzing training data.
Zoo’s AI text-to-CAD platform
Central to Zoo’s digital design ecosystem is its free-to-access, open-source Text-to-CAD platform that can convert short, text prompts into 3D CAD models. These can be exported as STL, PLY, OBJ, STEP, GTLF, GLB, and FBX files, ready for 3D printing or editing on external CAD software.
We tested the CAD generation tool, and achieved mixed results. In its current form, the platform requires users to avoid prompting vague and nebulous concepts. It is also limited to single objects, and assemblies cannot yet be handled. Users are encouraged to describe objects that can be represented in geometric shapes and to be “as explicit as possible” by clearly outlining size, shape, and feature requirements.
These requirements were confirmed when we gave vague prompts like “Make a classic 3D benchy,” “Make a front brake disk air scoop for a motorbike,” and “Steering wheel for a car.” These models either came out incredibly simplistically or failed to resemble the target shape altogether. It is worth recalling that generative AI image-based tools produced an inordinate amount of far-from-photo-realistic imagery when initially breaking through into wider public consciousness; AI text-to-CAD may chart a similar course.
However, we achieved improved results when refining the prompts to include more specific details. For instance, “Make a 10cm diameter helical gear with 20 teeth” produced a much better part that could be functional if 3D printed.
Zoo’s machine learning models are reportedly fine-tuned on user input. According to Zoo, this allows designers to leverage existing data and create specialized Text-to-CAD generators without having to build or maintain their own infrastructure. Looking to the future, the tech startup plans to grant users the ability to fine-tune Text-to-CAD based on their proprietary data sets for exclusive use.
AI-powered 3D printing software
Text-to-CAD tools are a growing theme in the additive manufacturing sector, allowing those with minimal digital design experience to create 3D printable models. Earlier this year, AI software startup Backflip made headlines after raising $30 million in funding. Founded by CEO Greg Mark and CTO David Benhaim, the new generative AI tool can convert images and text prompts into 3D printable parts.
Backed by venture capital firms NEA and Andreessen Horowitz (a16z), the new platform reportedly allows designers to create high-resolution models of complex designs in minutes. Once generated, the files can be 3D printed in metal, carbon fiber, or plastic. According to Mark, Backflip represents a “giant leap forward in bringing design and manufacturing back to the US.”
This followed the launch of PrintMon Maker, Bambu Lab’s AI-powered 3D model generator. Available through MakerWorld, the free tool allows users to create 3D printable Pokemon-style characters from text and image prompts. The tool is integrated into Bambu Lab’s Bambu Studio software. As such, generated models can be imported directly into the slicer without adjustments, optimizing them for multi-color 3D printing.
Elsewhere, 3D printing materials specialist polySpectra offers neThing.xyz. This AI-powered generative design tool allows users to create 3D printable CAD files using straightforward prompts. The tool seeks to make 3D CAD creation more accessible, lowering the barrier of entry to additive manufacturing in the process.
Who won the 2024 3D Printing Industry Awards?
Subscribe to the 3D Printing Industry newsletter to keep up with the latest 3D printing news.
You can also follow us on LinkedIn, and subscribe to the 3D Printing Industry Youtube channel to access more exclusive content.
Featured image shows a helical gear CAD model generated using Zoo’s text-to-CAD tool. Image by 3D Printing Industry.
