Comparing Text-to-Image Models for Nitik Batik Motifs with LLM-based Prompt Enhancement

Galih Gunardi, Muhammad Iqbal Akbar

Abstract


Pretrained text-to-image models enable batik motif generation without fine-tuning. However, the performance of general-purpose models and the benefits of Large Language Model (LLM)-based prompt enhancement for batik motif generation have not yet been systematically investigated. This study compares Stable Diffusion 1.5 (SD 1.5) and SDXL in generating Yogyakarta Nitik Batik motifs under zero-shot conditions and evaluates the effects of LLM-based prompt enhancement. Enhanced prompts were generated using Gemini through a single zero-shot instruction, without few-shot examples or system instructions. A quantitative experimental approach was employed, involving four experimental configurations (two models and two prompt types) across 15 motifs from the Batik Nitik 960 dataset, with 25 images generated per motif using fixed random seeds. Evaluation was conducted using the Fréchet Inception Distance (FID), calculated against original images from the same 15 motifs, and the CLIP Score, measured against a shared reference text. Paired t-tests were performed at the motif level. SD 1.5 with enhanced prompts achieved the lowest FID score (296.714); however, the differences among closely ranked configurations could not be established with sufficient certainty. Using the same reference text, prompt enhancement did not improve CLIP Scores (SD 1.5: −0.40; SDXL: −1.61). Its effect on FID was model-dependent, improving the results for SD 1.5 while worsening them for SDXL. All enhanced prompts exceeded the 77-token limit and were consequently truncated.

Keywords


batik nitik; large language model; prompt enhancement; stable diffusion; text-to-image

Full Text:

PDF

References


R. D. I. Wahid, N. Yudistira, C. Dewi, I. N. Sari, D. Pradhikta, and Fatmawati, "Prompt Conditioned Batik Pattern Generation using LoRA Weighted Diffusion Model with Classifier-Free Guidance," IEEE Access, Vol. 13, pp. 2436–2448, 2025, DOI: 10.1109/ACCESS.2024.3523494.

I. K. A. Purnawan and I. K. G. J. Suparjana, "Fine-Tuning Text-to-Image Diffusion Models with LoRA for Generative Synthesis of Balinese Endek Textile Patterns," Jurnal Nasional Pendidikan Teknik Informatika: JANAPATI, Vol. 15, No. 1, pp. 46–65, 2026, DOI: 10.23887/janapati.v15i1.111817.

Q. Hu, Y. Peng, J. Xu, Z. Shao, Z. Tian, and J. Chen, "Adaptive Stylized Image Generation for Traditional Miao Batik using Style-Conditioned LCM-LoRA Enhanced Diffusion Models," Mathematics, Vol. 13, No. 12, 2025, DOI: 10.3390/math13121947.

O. Octadion, N. Yudistira, and D. Kurnianingtyas, "Synthesis of Batik Motifs using a Diffusion-Generative Adversarial Network," Multimedia Tools and Applications, Vol. 84, pp. 3407–3438, 2025, Doi: 10.1007/s11042-025-20620-9.

W. Li, J. Wang, and X. Zhang, "PROMPTIST: Automated Prompt Optimization for Text-to-Image Synthesis," in Natural Language Processing and Chinese Computing, NLPCC 2024, LNCS Vol. 15360, Springer, Singapore, 2025, DOI: 10.1007/978-981-97-9434-8_23.

R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, "High-Resolution Image Synthesis with Latent Diffusion Models," in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2022, pp. 10684–10695.

E. J. Hu et al., "LoRA: Low-Rank Adaptation of Large Language Models," in Proc. Int. Conf. Learn. Represent. (ICLR), 2022.

D. Podell et al., "SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis," in Proc. Int. Conf. Learn. Represent. (ICLR), 2024.

A. Radford et al., "Learning Transferable Visual Models from Natural Language Supervision," in Proc. 38th Int. Conf. Mach. Learn. (ICML), PMLR Vol. 139, 2021, pp. 8748–8763.

S. Jayasumana, S. Ramalingam, A. Veit, D. Glasner, A. Chakrabarti, and S. Kumar, "Rethinking FID: Towards a Better Evaluation Metric for Image Generation," in Proc. IEEE/CVF CVPR, 2024, pp. 9307–9315, DOI: 10.1109/CVPR52733.2024.00889.

H. Gani, S. F. Bhat, M. Naseer, S. Khan, and P. Wonka, "LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts," in Proc. ICLR, 2024.

A. E. Minarno, H. A. Nugroho, and I. Soesanti, "Batik Nitik 960," Mendeley Data, V3, 2023, DOI: 10.17632/sgh484jxzy.3.

M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, "GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium," in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), Vol. 30, 2017.

Hugging Face, "Diffusers: State-of-the-Art Diffusion Models," GitHub repository, 2022. [Online]. Available: https://github.com/huggingface/diffusers

Google DeepMind, "Gemini: A Family of Highly Capable Multimodal Models," arXiv:2312.11805, 2023.

J. Hessel, A. Holtzman, M. Forbes, R. Le Bras, and Y. Choi, "CLIPScore: A Reference-Free Evaluation Metric for Image Captioning," in Proc. 2021 Conf. Empirical Methods in Natural Language Processing (EMNLP), 2021, pp. 7514–7528, DOI: 10.18653/v1/2021.emnlp-main.595.

N. Carlini et al., "Extracting Training Data from Diffusion Models," in Proc. 32nd USENIX Security Symposium, 2023, arXiv:2301.13188.

S. Nayak, M. Bhatia, X. Zhang, V. Rieser, L. A. Hendricks, S. van Steenkiste, Y. Goyal, K. Stańczak, and A. Agrawal, "CulturalFrames: Assessing Cultural Expectation Alignment in Text-to-Image Models and Evaluation Metrics," in Findings of the Association for Computational Linguistics: EMNLP 2025, 2025, pp. 20918–20953.

K. Arzomand, T. Kalganova, and M. J. F. Rustell, "HARF: A Human–AI Collaborative Framework for Cultural Heritage Reconstruction with Expert-Guided Multi-Platform Generative AI and Systematic Prompt Engineering," Digital Applications in Archaeology and Cultural Heritage, Vol. 42, e00554, 2026, DOI: 10.1016/j.daach.2026.e00554.

A. E. Minarno, I. Soesanti, and H. A. Nugroho, “Batik Nitik 960 Dataset for Classification, Retrieval, and Generator,” Data, Vol. 8, No. 4, Art. No. 63, 2023, DOI: 10.3390/data8040063.




DOI: https://doi.org/10.32520/stmsi.v15i11.7020

Article Metrics

Abstract view : 0 times
PDF - 0 times

Refbacks

  • There are currently no refbacks.


Creative Commons License
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.