Text2CAD is an innovative AI framework allowing designers to create parametric CAD models from text prompts, whether they are abstract descriptions or detailed instructions. The framework includes a unique Data Annotation Pipeline that utilizes open-source LLMs and VLMs to create varying complexities and details in the text prompts. The Text2CAD Transformer uses a Transformer-based architecture to generate CAD designs from input text prompts. Visual examples showcase the diverse results achieved with different prompts, such as generating a ring-like model without explicitly mentioning ‘ring.’ Quantitative evaluations demonstrate the effectiveness of Text2CAD compared to DeepCAD, GPT-4, and human evaluation methods.
https://sadilkhan.github.io/text2cad-project/