Revolutionary Waves: Shaping the Future of Manufacturing and Design with AI-Powered Text-to-CAD

The Wave of Revolution: The Future of Manufacturing and Design Shaped by AI "Text to CAD"

Introduction

In the manufacturing and design fields, the rapid development of generative AI is creating a new revolution. From Midjourney and DALL-E generating images from text to GitHub Copilot generating code from text, the scope of generative AI applications is expanding daily. Following these technological advancements, a technology called "Text to CAD" has been attracting attention in recent years.

Text to CAD is an innovative technology that analyzes instructions or design requirements in natural language (text) and automatically converts them into CAD data. For example, with a simple instruction like "a table 50cm wide and 30cm high," CAD data for a manufacturable 3D model is generated within minutes.

Compared to conventional general 3D generative AI, the biggest difference lies in the ability to directly generate "editable, precise CAD data." Many AIs that generate 3D from text create visually beautiful models, but they are usually in a format called "mesh," which has significant limitations in precision and editability for use in actual manufacturing or detailed design. On the other hand, the latest Text to CAD technology is making it possible to generate high-quality models that can be directly edited in existing CAD software.

Concept of Text to CAD: Flow from text prompt to CAD data generation

In this article, we will explain in detail the latest technological trends in Text to CAD, major research cases, actual PoCs (Proof of Concept), and its impact on the manufacturing and design industries.

Technical Explanation of Text to CAD

Importance of the B-Rep Method

The core of the latest Text to CAD technology is the "B-Rep (Boundary Representation)" method. This is a fundamentally different approach from the "mesh" method commonly used in traditional 3D modeling.

B-Rep is a method of defining 3D objects using surfaces (boundary surfaces), precisely representing shapes through elements such as vertices, edges, and faces, and their positional relationships. In contrast, the mesh method represents shapes approximately as a collection of polygons.

Comparison between B-Rep and Mesh methods: Differences in editability and precision

The biggest differences between the two are as follows:

1. Precision and Editability: B-Rep allows for mathematically accurate representation and is easy to edit in CAD software at any level of precision. On the other hand, mesh is an approximate representation, making complex shape editing or dimension changes difficult.

2. Manufacturability: B-Rep has high affinity with actual manufacturing processes and retains detailed information necessary for manufacturing, such as drilling and fillets (rounding corners). It is difficult to retain manufacturing information in a mesh.

3. File Formats: B-Rep can be saved in standard CAD formats such as STEP and IGES, and is compatible with many industrial CAD software programs.

Latest research, such as Zoo's Text-to-CAD tool, has developed technologies that can directly generate this B-Rep, contributing significantly to practical CAD data generation.

Fusion of Large Language Models (LLMs) and Visual Recognition AI

Another core aspect of Text to CAD technology is the fusion of Large Language Models (LLMs) and image recognition AI. Particularly noteworthy is a framework called "CADFusion" developed by Chinese researchers, which uses an LLM as a backbone and adopts an approach that alternates between sequential learning stages and visual feedback stages.

In this method, the LLM first analyzes text instructions and generates a CAD operation sequence. Next, it actually renders a 3D model from that operation sequence and utilizes visual quality as feedback for learning. This allows the model to learn by considering both text instructions and visual output results, enabling higher-quality CAD data generation.

How CAD Sequence Generation Works

The actual CAD model generation process is not a simple "text input → 3D model output." In current cutting-edge research, the mainstream approach is to convert text instructions into a sequence of step-by-step CAD operations.

For example, when a text instruction like "a sphere with a radius of 5cm on top of a cylinder 10cm high" is received, the system converts it into an operation sequence like the following:

1. Draw a circle on the XY plane (radius: 5cm)
2. Extrude the circle (height: 10cm)
3. Create a point at the center of the top surface of the cylinder
4. Create a sphere centered at that point (radius: 5cm)

Transformer-based architectures like "Text2CAD" are used to generate such operation sequences, trained to handle text instructions of varying difficulty levels. By outputting these operation sequences in a format executable by existing CAD engines, practical CAD data is generated.

Process flow of Text to CAD: From text analysis to CAD generation

Latest Research and Key Technologies

In the field of Text to CAD, several significant research and development projects have been advancing in recent years. Here, we introduce some particularly noteworthy technologies and research results.

Zoo's Open-Source Text-to-CAD Technology

Text-to-CAD, developed by the American startup "Zoo," is one of the most advanced technologies in this field in terms of practical application. The company released an alpha version in late 2023 and provides a web interface that can generate B-Rep format CAD data (STEP format) directly from text prompts.

The features of Zoo's system are as follows:

1. Open Source: The Text-to-CAD tool is available for anyone to use for free, with plans to allow companies to fine-tune it using their own datasets in the future.

2. Practical Output Formats: It supports a wide range of file formats, including STL, PLY, OBJ, STEP, GTLF, and GLB, ensuring compatibility with major 3D printers and CAD software.

3. Scalability via API: Through the company's Design API and Machine Learning API, it provides a mechanism to programmatically generate and manipulate CAD files.

This tool is especially well-suited for rapid prototyping in manufacturing and product design, holding the potential to significantly improve the efficiency of trial and error during the early stages of design.

You can try Zoo's Text-to-CAD on their official website (https://zoo.dev/text-to-cad).

CADFusion Framework

"CADFusion," announced in early 2025 by a research team in China, proposes a new approach for generating CAD models from text instructions. The most significant feature of this research lies in its two-stage learning process that combines sequential learning and visual feedback.

To briefly explain the CADFusion approach:

1. Sequential Learning Stage: Learning is performed to generate CAD parametric sequences from text instructions using a Large Language Model (LLM).

2. Visual Feedback Stage: The generated CAD sequence is rendered, and its visual quality is evaluated. By giving rewards for sequences that produce good models and penalties for those that do not, the model's learning is encouraged.

The innovation of this method is that it considers not only the logical consistency between the text instruction and the CAD sequence but also the quality of the final visual output. The research team reported that this two-stage approach resulted in significantly improved quality compared to conventional methods.

Text2CAD: Challenging Parametric CAD Model Generation

Announced in late 2024 by DFKI (German Research Center for Artificial Intelligence), "Text2CAD" proposes a framework capable of generating CAD models from a variety of text instructions, ranging from beginner to expert levels.

The main features of Text2CAD are:

1. Step-by-step Difficulty Support: It is designed to handle text instructions of various difficulty levels, from simple instructions ("Generate two concentric cylinders") to complex ones including detailed dimensions and shape characteristics.

2. Data Annotation Pipeline: They have developed a mechanism to automatically generate datasets consisting of pairs of text instructions and CAD operation sequences using the latest AI models such as Mistral-50B and LLaVA-NeXT.

3. End-to-End Architecture: A transformer-based model has been built to directly convert text instructions into CAD operation sequences, outputting them in a format executable by existing CAD software.

This research holds great potential, particularly for supporting CAD beginners and improving the efficiency of the design process, and is attracting attention as a technology that will accelerate the democratization of CAD.

These research projects are evolving while influencing each other, and the accuracy and practicality of generating CAD data from text are improving daily. Especially now that actual applications in the industry are beginning, Text to CAD technology holds the potential to innovate the entire manufacturing and design process.

Demonstration Experiments and Case Studies

Text to CAD technology is gradually moving from the research stage to the practical stage, and various demonstration experiments and advanced use cases are appearing around the world. Here, we introduce some particularly noteworthy examples.

Demonstration Experiments in Architecture and Design

The field of architecture and design is one of the areas that could benefit significantly from Text to CAD technology.

Automatic Generation of Residential Floor Plans by "Maket"

In the early stages of architectural design, it is necessary to quickly create multiple floor plan options based on client requirements. A service called "Maket" provides a feature that automatically generates residential floor plans when simple requirements are entered as text, allowing for immediate download in CAD data (DXF format).

Starting from relatively simple requirements such as "for a family of four, spacious living and dining area, with a home office," the system generates multiple feasible floor plan proposals that also consider building codes by interactively refining details. Professionals can use these proposals as a starting point to proceed with their designs efficiently.

Maket provides an AI-driven architectural design platform on its official website (https://maket.ai/). In Japan, Lib Work Co., Ltd. has partnered with Maket to launch the "Generative AI Housing" project.

Obayashi Corporation's AI Design Support Tool

Obayashi Corporation, a major Japanese general contractor, has internally developed an AI tool called "AiCorb" that generates multiple exterior design proposals from sketches or simple 3D models. This can be considered a derivative application of Text to CAD, where it takes simple sketches as input, in addition to text instructions, and generates developed 3D architectural models.

In this initiative, the emphasis is placed on supporting designers to quickly explore various design possibilities rather than restricting their creativity. The generated models serve as a starting point for designers to further develop into more detailed designs.

Details about Obayashi Corporation's "AiCorb" are introduced on their official website (https://www.obayashi.co.jp/news/detail/news20220301_3.html).

Examples of Application in Industrial Product Design

In the manufacturing and product design fields, advanced initiatives utilizing Text to CAD technology are also underway.

Rapid Prototyping at an Automotive Parts Manufacturer

A major automotive parts manufacturer is working to improve the efficiency of brainstorming during the initial design phase by utilizing Zoo's Text-to-CAD. For example, they have established a workflow where they can immediately generate multiple design proposals from requirements like "a lightweight side mirror cover with enhanced aerodynamic performance," which designers then use as a base for more detailed studies.

Through this method, the creation of initial design proposals, which previously took about a week, has been reduced to a few hours, making it possible to explore a wider range of design options. Ultimately, fine adjustments and manufacturability reviews by human designers are still necessary, but the transition from brainstorming to detailed design has been significantly streamlined.

Example of a Product Design Studio

A product design studio in Denmark is utilizing Text to CAD to visualize product concepts in real-time during initial client meetings. By inputting client requirements as text on the spot and displaying them as 3D models, they are resolving discrepancies in understanding early on and accelerating the overall design process.

This approach is particularly effective in smoothing communication with clients who lack technical design expertise, contributing significantly to shared understanding during the early stages of a project.

PoC Case Studies by Startups

Emerging companies and startups are utilizing Text to CAD technology with more innovative and experimental approaches.

Link AI's Architectural CAD Support System

Link AI, a Japanese startup, is conducting demonstration experiments for a service that customizes technology for generating CAD data from text for the architecture and housing industry. In this system, when architectural requirements are entered as text, it generates a feasible 3D model while considering laws and regulations such as the Building Standards Act and the Fire Service Act.

Furthermore, their system does not just generate models; it also features a function to automatically detect areas that may violate regulations and propose corrections. This helps reduce the risk of design errors and regulatory violations.

Details about Link AI can be found on their official website (https://www.linkai.co.jp/), where they are actively working on developing AI solutions for the construction industry.

Automation of Custom Part Design

Another startup for the manufacturing industry is developing a system that automates the design of custom parts by combining a company's existing CAD data with natural language modification instructions. For example, with instructions like "increase the thickness of this part by 2mm and add a 5mm chamfer to the corners," a CAD model with quick modifications to an existing part can be generated.

This approach is especially effective for designing part families with many similar variations, contributing to improved responsiveness for product customization.

These cases show that Text to CAD technology is moving beyond the mere research stage and beginning to bring value to actual business processes. While it has not yet reached the level of completely replacing human designers, it can be said that it is contributing significantly to the efficiency of the design process and the expansion of creative possibilities.

Impact on Industry and Future Outlook

Text to CAD technology holds the potential to bring about fundamental transformations to the manufacturing and design fields. Here, we consider the impact of this technology on the industry and its future outlook.

Democratization and Efficiency of the Design Process

One of the most significant impacts of Text to CAD technology is the "democratization" of the design process. Traditionally, operating CAD required specialized training and experience, but being able to instruct models using natural language will allow people without CAD expertise to create 3D models.

Lowering the Barrier from Idea to Shape

In product development, quickly visualizing shapes in the initial stages of an idea is crucial. Since Text to CAD can convert vague ideas like "I want to make something like this" into concrete 3D models within minutes, it can significantly shorten the cycle of idea validation.

For example, if marketing managers or executives could directly shape their product concepts without going through designers, communication within the organization would become smoother, accelerating consensus-building on the direction during the early stages of product development.

Dramatic Reduction in Design Time

In corporate design departments, Text to CAD will significantly reduce the time required to generate initial design proposals. As seen in the previously mentioned automotive parts manufacturer case, work that conventionally took a week could potentially be shortened to a few hours. Additionally, the time spent on designing variations can be drastically reduced.

This will allow designers to focus on more creative design challenges and optimization, which is expected to lead to greater product innovation and quality improvements.

Potential as a Driver for Manufacturing DX

In the digital transformation (DX) of the manufacturing industry, Text to CAD has the potential to become a major driving force.

Promoting Data-Driven Design

By learning from past design and product data, Text to CAD systems will be able to generate more practical and manufacturable models. This will allow companies to incorporate their design know-how into AI and promote data-driven design approaches.

For example, by training the system with a combination of past design data and performance evaluation data, it may become possible to automatically generate optimal design proposals based on past success stories from performance requirements such as "a high-efficiency heat exchanger."

Integration with Supply Chain Optimization

By linking Text to CAD with material procurement and production planning systems, there is potential to further streamline processes from design to manufacturing. For instance, it could generate design proposals that consider constraints on available parts and materials, or automate design optimization that accounts for manufacturing costs and lead times.

In this way, Text to CAD technology is poised to become one of the core technologies of DX in the manufacturing industry, potentially driving transformation across the entire design and manufacturing process.

Shifting Roles of Specialized Engineers and New Possibilities

With the spread of Text to CAD, the roles of CAD designers and engineers are expected to change.

Evolution of the Designer's Role

As the CAD operation itself becomes automated, the designer's work will shift from "creating models using CAD" to "evaluating and refining models generated by AI." Additionally, they will be able to focus on more advanced design challenges, such as complex functional optimization or the development of new design methodologies.

This is not a simple scenario of "AI taking away designers' jobs," but rather means that the role of the designer will shift toward high-value-added areas.

Demand for New Skill Sets

In the era of Text to CAD, the following skills will become more important than traditional CAD operation skills:

1. Prompt Engineering: The skill to effectively instruct AI to generate desired models.
2. Design Evaluation and Optimization: The skill to evaluate and improve the performance and manufacturability of generated models.
3. Domain Knowledge and Creativity: Deep knowledge of a specific industry and creative problem-solving abilities.

As a result, CAD education and training programs for designers are also expected to change.

The impact of Text to CAD technology in manufacturing and design, much like the influence of AI in other industries, will be both destructive and generative of new possibilities. Companies that can adopt this technology early and utilize it strategically are likely to establish a competitive advantage in the future manufacturing and design fields.

Realistic Steps for Implementation

How should companies and professionals interested in Text to CAD technology proceed with its implementation? Here, we explain realistic steps for implementation and points to consider.

Current Technical Constraints and Challenges

When considering the introduction of Text to CAD technology, it is important to understand the current technical constraints.

Limits of Precision and Complexity

Current Text to CAD technology shows high accuracy for generating relatively simple shapes and standard parts, but there are limits regarding complex mechanisms or special shapes. Specifically, the following challenges exist:

• High-precision tolerance specification: Detailed requirements such as strict tolerance specifications necessary for manufacturing are not yet sufficiently addressed.
• Special manufacturing requirements: Insufficient handling of specialized processing requirements, such as draft angles for injection molding.
• Conformity to industry-specific standards: Automatic conformity to specific industry standards and norms has not yet been realized.

For this reason, it is wise to start implementation in areas where accuracy requirements are not strict, such as relatively simple parts or initial concept designs.

Limits of Training Data

The accuracy of Text to CAD systems heavily depends on the data used for training. Current models are trained on general CAD data, but additional learning is required to adapt to specific industries or company-specific design styles.

Furthermore, to support domain-specific terminology or tacit knowledge, fine-tuning using existing in-house CAD data and text descriptions will be necessary.

Preparations for Implementation Companies Should Consider

Companies considering the introduction of Text to CAD technology are recommended to take a step-by-step approach as follows:

1. Setting Up Pilot Projects

It is important to start with small-scale pilot projects by selecting specific part types or design processes. For example:

• Designing simple variations of existing products
• Generating initial design proposals for standard parts
• Shape exploration in the conceptual design phase

Choosing areas where the risk of failure is low and results are easy to measure makes it easier to verify the effectiveness of the technology's introduction.

2. Organizing Internal Data

To effectively utilize Text to CAD technology, it is necessary to organize your company's design data into a usable form:

• Digitizing and organizing past CAD data and design specifications
• Creating paired data of CAD files and text descriptions
• Documenting design know-how

This data can be used for fine-tuning proprietary models in the future, and it also serves as knowledge for providing clear instructions to the AI.

3. Building Hybrid Workflows

When integrating Text to CAD into existing design processes, it is more realistic to build a hybrid workflow between human designers and AI rather than aiming for full automation:

• Initial design proposal generation by AI → Evaluation and revision by humans
• Conceptual sketching by humans → 3D modeling by AI → Detailed design by humans
• Generation of multiple options by AI → Selection of the optimal plan and finalization by humans

Such a hybrid approach allows you to leverage the creativity and efficiency of AI while combining it with human expertise and judgment.

Perspective on Human Resource Development

Preparing the human side is just as important as the technical side when introducing Text to CAD technology.

Skills for Effective Collaboration with AI

For designers to effectively utilize Text to CAD tools, the development of the following skills is necessary:

• Effective Prompt Creation: Verbalization skills to give appropriate instructions to the AI.
• Evaluation and Editing of Generative Models: The ability to evaluate models generated by AI and edit them efficiently when necessary.
• Understanding AI Characteristics: The judgment to understand the strengths and weaknesses of AI and assign tasks appropriately.

These skills will become indispensable for designers in the future.

Mindset Reform and Change Management

Organizational changes accompanying the introduction of new technology are also important considerations:

• Addressing anxiety regarding changes in the designer's role
• Mindset reform to position AI tools as partners rather than just assistants
• Reviewing evaluation and reward systems for new workflows

Especially for veteran designers who have used CAD for many years, it is necessary to emphasize that Text to CAD is not something that threatens their experience or knowledge, but rather a tool that amplifies them, in order to reduce resistance.

Text to CAD technology is still in its infancy and is not yet a fully mature technology. However, by understanding the characteristics of this technology now and proceeding with experimental introductions and preparations, companies will be able to secure a competitive advantage for the future.

Summary

In this article, we have explained the latest trends in "Text to CAD" technology, which generates CAD models from text, and its impact on the manufacturing and design fields. Here, we will summarize the content so far and present the future outlook.

Evolution Forecast for Text to CAD Technology

Text to CAD technology continues to evolve rapidly, and the following developments are expected over the next few years:

1. Improved Precision and Complexity: The technology will evolve to handle more complex shapes and specialized design requirements, expanding the scope of practical design work.

2. Emergence of Domain-Specific Models: Text to CAD models specialized for specific industries such as architecture, mechanical parts, and electronic equipment will emerge, enabling high-precision model generation that corresponds to the terminology and tacit knowledge of each field.

3. Support for Multimodal Inputs: Hybrid CAD generation systems that combine various input formats, such as handwritten sketches, voice instructions, and existing CAD models, in addition to text, are expected to develop.

4. Deeper Integration with Design Workflows: Moving beyond standalone tools, integration with PLM (Product Lifecycle Management) systems and CAE (Computer-Aided Engineering) tools will progress, becoming part of an ecosystem that supports the entire process from design to verification and manufacturing.

Opportunities and Challenges for Japanese Companies

In the Japanese manufacturing and design sectors, Text to CAD technology brings several unique opportunities and challenges:

Opportunities

1. Acceleration of Technology Transfer: In Japanese manufacturing sites with an aging workforce, there is potential to use this as a means to capture the tacit knowledge of skilled technicians into AI and pass it on to the next generation.

2. Addressing Labor Shortages: Partial automation of design work could allow fewer human resources to handle more design variations, potentially contributing to the mitigation of labor shortages.

3. Focus on High-Value-Added Design: Automation of routine work will make it possible to concentrate human resources on high-quality, high-value-added design, which is a strength of Japanese industry.

Challenges

1. Conservatism Toward Implementation: Japanese companies tend to emphasize proven methods and existing processes, and are cautious about introducing innovative AI technologies. How to strike this balance is a significant challenge.

2. Delays in Data Preparation: Many Japanese companies face challenges where past design data is not digitized or standardized, necessitating data preparation required for AI learning.

3. Need for Human Resource Development: There is an urgent need to train engineers who can utilize AI technology, and the re-education of existing designers and the hiring of talent with new skill sets will be a challenge.

Final Outlook

Text to CAD technology has the potential to bring about a fundamental transformation of the design and manufacturing process, comparable to the birth of CAD or the emergence of CAE. However, this technology will not replace human designers; instead, it will be a powerful tool that significantly enhances their creativity and productivity.

In the future, the role of designers is expected to shift from "people who operate CAD" to "people who clearly express design intent and collaborate with AI to derive optimal solutions." Companies that can adapt to this change and actively embrace new technologies will be the ones leading the next generation of manufacturing.

The evolution of Text to CAD technology has only just begun. With continued research and development and the accumulation of demonstration cases, the future of manufacturing and design will undoubtedly change even more significantly. Riding this wave of transformation and exploring new possibilities is what is required of the manufacturing industry from now on.

References

1. Zoo Text-to-CAD Official Documentation (2023) https://zoo.dev/text-to-cad
2. "Text-to-CAD Generation Through Infusing Visual Feedback in Large Language Models" (2025) arXiv
3. "Text2CAD: Generating Sequential CAD Models from Beginner-to-Expert Level Text Prompts" (2024) DFKI
4. Impact of AI on the Construction Industry: Case Study by Maket (2024) https://maket.ai/
5. "Creating CAD Models with Just Text! Latest Technology for Small Businesses" AI WAVE (2025)
6. Obayashi Corporation AI Design Support Tool "AiCorb" Official Information https://www.obayashi.co.jp/news/detail/news20220301_3.html
7. Link AI Co., Ltd. Official Website https://www.linkai.co.jp/