Improving Text-to-Image Consistency via Automatic Prompt Optimization
This work addresses reliability issues in text-to-image generation for users needing accurate visual outputs, though it is incremental as it builds on existing models with a novel prompting approach.
The paper tackles the problem of text-to-image models failing to produce images consistent with input prompts, such as object quantities and attributes, by introducing OPT2I, a framework that uses a large language model to iteratively optimize prompts, resulting in up to a 24.9% improvement in consistency scores on datasets like MSCOCO and PartiPrompts while maintaining image quality.
Impressive advances in text-to-image (T2I) generative models have yielded a plethora of high performing models which are able to generate aesthetically appealing, photorealistic images. Despite the progress, these models still struggle to produce images that are consistent with the input prompt, oftentimes failing to capture object quantities, relations and attributes properly. Existing solutions to improve prompt-image consistency suffer from the following challenges: (1) they oftentimes require model fine-tuning, (2) they only focus on nearby prompt samples, and (3) they are affected by unfavorable trade-offs among image quality, representation diversity, and prompt-image consistency. In this paper, we address these challenges and introduce a T2I optimization-by-prompting framework, OPT2I, which leverages a large language model (LLM) to improve prompt-image consistency in T2I models. Our framework starts from a user prompt and iteratively generates revised prompts with the goal of maximizing a consistency score. Our extensive validation on two datasets, MSCOCO and PartiPrompts, shows that OPT2I can boost the initial consistency score by up to 24.9% in terms of DSG score while preserving the FID and increasing the recall between generated and real data. Our work paves the way toward building more reliable and robust T2I systems by harnessing the power of LLMs.