EPIC: Efficient Predicate-Guided Inference-Time Control for Compositional Text-to-Image Generation

TL;DR

EPIC is a training-free, inference-time refinement framework for compositional text-to-image generation that improves accuracy and efficiency by predicate-guided search and targeted editing.

Contribution

EPIC introduces a novel predicate-guided search method for inference-time refinement in compositional T2I generation, significantly enhancing accuracy and reducing computational costs.

Findings

Prompt-level accuracy increased from 34.16% to 71.46%.

EPIC outperforms prior refinement baselines by 19.23 points.

Reduces image-model executions by 31%, MLLM calls by 72%, and tokens by 81%.

Abstract

Recent text-to-image (T2I) generators can synthesize realistic images, but still struggle with compositional prompts involving multiple objects, counts, attributes, and relations. We introduce EPIC (Efficient Predicate-Guided Inference-Time Control), a training-free inference-time refinement framework for compositional T2I generation. EPIC casts refinement as predicate-guided search: it parses the original prompt once into a fixed visual program of object variables and typed predicates, covering checkable conditions such as object presence, counts, attributes, and relations. Each generated or edited image is verified against this program using visual evidence extracted from that image. An image is judged to satisfy the prompt only when all predicates are satisfied; otherwise, failed predicates decide the next step, routing local failures to targeted editing and global failures to resampling while the fixed visual program remains unchanged. On GenEval2, EPIC improves prompt-level accuracy from 34.16% for single-pass generation with the base generator to 71.46%. Under the same generator/editor setting and maximum image-model execution budget, EPIC outperforms the strongest prior refinement baseline by 19.23 points while reducing realized cost by 31% in image-model executions, 72% in MLLM calls, and 81% in MLLM tokens per prompt.