TEXT TO IMAGE GENERATOR USING DALL-E API FOR YORUBA PROMPTS

This project focuses on the development of a text-to-image generator capable of processing Yoruba language prompts and producing relevant images using the DALL·E API. The objective was to address the limitation of existing generative models, which primarily function with English inputs, by creating a system that allows Yoruba speakers to express prompts in their native language. To achieve this, a dataset of Yoruba–English parallel texts was collected, preprocessed, and split for training, validation, and testing. A Yoruba-to-English translation model was then trained from scratch using TensorFlow. The translated outputs were interfaced with the DALL·E API to generate images. Evaluation was carried out using BLEU and ChrF scores to assess translation quality, alongside both automated and human assessments of image relevance and accuracy. Results demonstrated that the system was able to successfully translate Yoruba prompts and generate corresponding images with reasonable accuracy, though challenges persisted with complex expressions and idiomatic phrases. In conclusion, this work highlights the potential of integrating indigenous African languages into generative AI systems, thereby promoting inclusivity in AI applications and paving the way for future improvements through larger datasets and model refinements.