Towards High-Fidelity CAD Generation via LLM-Driven Program Generation and Text-Based B-Rep Primitive Grounding

Jiahao Li; Qingwang Zhang; Qiuyu Chen; Guozhan Qiu; Yunzhong Lou; Xiangdong Zhou

cs.CVMar 12, 2026

Towards High-Fidelity CAD Generation via LLM-Driven Program Generation and Text-Based B-Rep Primitive Grounding

Authors: Jiahao Li, Qingwang Zhang, Qiuyu Chen, Guozhan Qiu, Yunzhong Lou, Xiangdong Zhou

Organizations: Fudan University · Shanghai Jiao Tong University

Abstract

The field of Computer-Aided Design (CAD) generation has made significant progress in recent years. Existing methods typically fall into two separate categories: parametric CAD modeling and direct boundary representation (B-Rep) synthesis. In modern feature-based CAD systems, parametric modeling and B-Rep are inherently intertwined, as advanced parametric operations (e.g., fillet and chamfer) require explicit selection of B-Rep geometric primitives, and the B-Rep itself is derived from parametric operations. Consequently, this paradigm gap remains a critical factor limiting AI-driven CAD modeling for complex industrial product design. This paper presents FutureCAD, a novel text-to-CAD framework that leverages large language models (LLMs) and a B-Rep grounding transformer (BRepGround) for high-fidelity CAD generation. Our method generates executable CadQuery scripts, and introduces a text-based query mechanism that enables the LLM to specify geometric selections via natural language, which BRepGround then grounds to the target primitives. To train our framework, we construct a new dataset comprising real-world CAD models. For the LLM, we apply supervised fine-tuning (SFT) to establish fundamental CAD generation capabilities, followed by reinforcement learning (RL) to improve generalization. Experiments show that FutureCAD achieves state-of-the-art CAD generation performance. Code and dataset are available at: https://github.com/JohanStackk/FutureCAD

Figures & tables

Figure 1: Given a textual description, FutureCAD synergizes LLM-driven program generation and text-based B-Rep primitive grounding to support feature-based CAD modeling, enabling high-fidelity CAD model generation.

Figure 2: Overview of the FutureCAD framework. Left: An example CadQuery program with text-based queries (highlighted in orange ) for B-Rep primitive selection. Top: The training pipeline includes BRepGround training for primitive grounding, and LLM training with supervised fine-tuning (SFT) followed by reinforcement learning (RL) with GSPO using Chamfer Distance-based rewards. Bottom: Illustration of the CAD modeling process: the LLM generates parametric features executed by the CAD kernel, and when operations require primitive references, the LLM produces textual queries for BRepGround to resolve, with the results enabling the kernel to proceed.

Figure 3: Architecture of BRepGround . (a) The B-Rep encoder extracts face and edge features and produces primitive embeddings via a GNN and adaptive layer. (b) The text encoder processes the query using pretrained BERT. (c) The fusion module fuses primitive and text embeddings through self-attention and cross-attention, followed by a classifier that predicts target primitives.

Methods	FutureCAD Dataset									Fusion 360 (Out-of-Distribution)
92pt.	Advanced					Standard				Fusion 360 (Out-of-Distribution)
	G -F1 $\uparrow$	R -F1 $\uparrow$	Median CD $\downarrow$	Mean CD $\downarrow$	IR $\downarrow$	G -F1 $\uparrow$	Median CD $\downarrow$	Mean CD $\downarrow$	IR $\downarrow$	G -F1 $\uparrow$	Median CD $\downarrow$	Mean CD $\downarrow$	IR $\downarrow$
CAD-Translator	-	-	-	-	-	79.08	94.23	152.60	3.57	76.82	159.98	211.72	3.86
Text2CAD	-	-	-	-	-	81.39	77.65	132.04	1.90	80.24	112.77	178.42	3.56
Text-to-CadQuery	-	-	-	-	-	87.04	109.83	173.69	8.38	84.36	188.88	232.56	11.70
cadrille	-	-	-	-	-	84.13	71.89	125.58	2.57	84.53	143.04	206.93	5.02

Table 1: Main results on text-to-CAD generation.

\dagger

indicates that we extend the command space of the method to support advanced features and retrain on our dataset, while other methods are evaluated only on the standard subset due to lack of such support. Results are reported on our FutureCAD, as well as the out-of-distribution Fusion360 test set. G -F1 covers generation commands ( sketch , extrude , revolve ); R -F1 covers refinement commands ( fillet , chamfer , shell ). CD is multiplied by

10^{3}

and all other metrics are multiplied by

100\%

Figure 4: Qualitative comparison of text-to-CAD generation. Left: Results on the advanced subset, which includes models with advanced operations. Right: Results on the standard subset. Our method produces more accurate CAD models aligned with input descriptions.

Methods	Recall@3 $\uparrow$	Recall@5 $\uparrow$	Recall@10 $\uparrow$	mAP $\uparrow$	F1 $\uparrow$
LLM-Query	-	-	-	-	21.51
CLIP-DE	51.81	62.57	75.01	57.53	31.05
Late Fusion	51.43	63.55	76.72	58.18	43.32
Ours	56.39	65.86	78.47	63.07	50.23

Table 2: Results on text-based B-Rep grounding task. LLM-Query prompts an LLM to generate code using CadQuery native selector APIs. CLIP-DE trains a dual encoder with CLIP-style contrastive loss. Late Fusion concatenates pooled text features with primitive embeddings for classification. LLM-Query produces binary outputs and is excluded from ranking metrics. All metrics are multiplied by 100%

Methods	G -F1 $\uparrow$	R -F1 $\uparrow$	Median CD $\downarrow$	Mean CD $\downarrow$	IR(%) $\downarrow$
SFT only	79.82	91.44	44.00	85.15	42.67
w/o BRepGround	80.36	82.22	38.50	76.01	8.29
Ours	80.54	85.35	31.12	71.39	1.01

Table 3: Ablation study for text-to-CAD generation on the advanced subset. w/o BRepGround replaces text-based primitive selection with point-based nearest neighbor retrieval.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 5: Feature length distribution comparison between DeepCAD and our FutureCAD dataset.

Figure 6: Comparison of DeepCAD and FutureCAD datasets. DeepCAD (left) contains simple geometric primitives, while FutureCAD (right) features complex real-world industrial parts with advanced operations such as fillet , chamfer , and shell .

Figure 7: Typical failure cases of our method. (a) Spatial Misalignment: incorrect positioning of geometric primitives. (b) Missing Geometric Details: omission of specified refinement features such as fillets. (c) Text-Model Inconsistency: generated geometry contradicts the spatial arrangement described in the input text.

Text-to-CAD Generation

Text-to-CAD generation turns natural-language design intent into executable, editable parametric 3D models and assemblies.

See all 48 papers

CADFS: A Big CAD Program Dataset and Framework for Computer-Aided Design with Large Language Models
May 3, 2026Vladislav Pyatov, Gleb Bobrovskikh, Saveliy Galochkin +6Parametric CAD Modeling Program Synthesis
1Applied AI Institute · 2AI Foundation and Algorithm Lab · 3AXXX +1
HierCAD: Hierarchical Text-to-CAD Design via Structure Alignment and Parameter Grounding
Jul 13, 2026Jimin Xu, Tianbao Wang, Tao Jin +1Parametric CAD Modeling Text-to-CAD Generation
College of Computer Science and Technology, Zhejiang University, Hangzhou, China
PR-CAD: Progressive Refinement for Unified Controllable and Faithful Text-to-CAD Generation with Large Language Models
Mar 27, 2026Jiyuan An, Jiachen Zhao, Fan Chen +6Text-to-CAD Generation Generative Modeling
National Language Resources Monitoring and Research Center for Print Media, Beijing Language and Culture University, China · School of Information Science, Beijing Language and Culture University, China · School of Computer Science and Technology, Beijing Jiaotong University, China +3

May 3, 2026J/K move · Enter open · S save

CADFS: A Big CAD Program Dataset and Framework for Computer-Aided Design with Large Language Models

Vladislav Pyatov, Gleb Bobrovskikh, Saveliy Galochkin, Nikita Boldyrev+5

1Applied AI Institute · 2AI Foundation and Algorithm Lab · 3AXXX+1

Parametric CAD Modeling Program Synthesis Text-to-CAD Generation

Open paper AskPDF arXiv

We introduce CADFS, a data-centric framework that enables large vision-language models to generate complex CAD design histories. Existing generative CAD systems are restricted to sketch-extrude operations due to simplified representations and limited datasets. We address this by introducing a FeatureScript-based representation and constructing a dataset of 450k real-world CAD models spanning 15 modeling operations. We obtain the dataset via a new pipeline that reconstructs clean, executable FeatureScript programs and provides multimodal annotations. Fine-tuning a VLM on this representation yields state-of-the-art results in text-conditioned CAD generation and image-based reconstruction, producing more accurate, diverse, and feature-rich designs than prior frameworks. Ablations show that each individual component of our framework, i.e., the FeatureScript representation, the extended operation set, and representation-aligned textual descriptions, significantly improves performance. Our framework substantially broadens the complexity and realism achievable in generative CAD. The CADFS framework and the new dataset are available at https://voyleg.github.io/cadfs/.

Abstract

Text-to-CAD Generation

Explore similar work

CADFS: A Big CAD Program Dataset and Framework for Computer-Aided Design with Large Language Models

HierCAD: Hierarchical Text-to-CAD Design via Structure Alignment and Parameter Grounding

PR-CAD: Progressive Refinement for Unified Controllable and Faithful Text-to-CAD Generation with Large Language Models