AutoSlice: A Fully Automatic LLM-driven Slicing Configuration Framework for Additive Manufacturing

Producing reliable, production-quality 3D prints depends on an expert tuning multiple slicer parameters, which is a manual and graphical interface-bound process. While Large Language Models (LLMs) excel at reasoning tasks, their application to automate the slicing workflow remains underexplored. To address this challenge, we developed AutoSlice, a novel, modular framework that automates the generation of slicer configurations, which has an LLM as the core reasoning engine. The framework is designed to be scalable with a plug-and-play architecture, taking advantage of future versions in LLMs. The AutoSlice pipeline locally extracts non-invertible geometric features from a 3D model. These non-invertible design features, combined with schema-anchored rules and printer-material metadata, are then passed to an LLM agent, which reasons about the geometry and machine-specific heuristics to generate a complete and syntactically valid configuration file. This framework executes a headless command-line slicer, yielding machine-ready G-code. AutoSlice encapsulates expert knowledge, demonstrating a viable path to democratizing AM by enabling non-experts to reliably translate digital designs into high-quality objects. We conducted a comprehensive evaluation to demonstrate the framework's robustness and modularity. We evaluated several versions of LLMs: both proprietary (Gemini-2.5-Flash, GPT-5-Mini) and open-source (Llama-3.2-3B-Instruct, Qwen-2.5-7B-Instruct), and evaluated also the Zero-Shot and One-Shot prompting strategies across two printers (Ender 3 S1, Prusa XL) and materials (PLA, ABS). The results show that while smaller models struggle with reliability, larger LLMs achieve high efficacy. The One-Shot strategy with proprietary LLMs yielded over 97% valid G-code generation and printability, validated by successful physical prints. Full Paper Available at https://www.sciencedirect.com/science/article/pii/S2213846326001355 Citation: @article{BINBASHAR2026698, title = {AutoSlice: A Fully Automatic LLM-driven Slicing Configuration Framework for Additive Manufacturing}, journal = {Manufacturing Letters}, volume = {49}, pages = {698-709}, year = {2026}, note = {54th SME North American Manufacturing Research Conference}, issn = {2213-8463}, doi = {https://doi.org/10.1016/j.mfglet.2026.07.084}, author = {Syed Ziaul {Bin Bashar} and Jesus Diaz and Chaitanya Mahajan and Satyajayant Misra}, keywords = {Additive Manufacturing, large language models, auto slicing, autonomous 3D printing, AM pipeline}, }