ImageRSI improves text-to-image generation using verification and repair
ImageRSI: Verification-Driven Recursive Self-Improvement for Agentic Image Generation
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Generating images from complex text requests can be tricky and often produces mistakes. The authors present ImageRSI, a system that keeps the image generator the same but uses a process to check and fix images, learn from past fixes, and improve its judging method over time. This approach helps the system produce better images that more closely match difficult requests. They also created a benchmark with hard image-generation challenges to test their method.
What this means in practice
- •For digital content creators: Generate more accurate images from complex text prompts by using a system that verifies and repairs images based on prior successful fixes.
- •For software development teams: Integrate verification-driven self-improvement techniques to enhance the reliability of AI-generated images in applications requiring detailed visual content.
Authors
Shuo Han, Zhilin Wang, Peng Ling, Wenjia Gong, Yang Liu, Guanjun Jiang, Yukun Cao, Xike Xie
Abstract
Complex text-to-image requests remain difficult to satisfy reliably. Verifying and repairing an image can fix an individual output, but a successful repair does not necessarily carry over to other requests. We present ImageRSI, a framework that keeps the image generator frozen and combines three components: five-dimensional verification and repair within each task, an experience library shared across tasks, and offline updating of the judge. Candidate experiences enter the library only after paired A/B validation, and the judge is updated with supervised fine-tuning and GRPO on accumulated image-judgment records and evaluated before reuse. We also introduce ImageReq-Bench, a benchmark of 961 requests built around hard cases in real-world image-generation requests, each paired with an evaluation checklist. ImageRSI achieves a Checklist Strict Accuracy (CSA) of 52.6% on MindBench (with Nano Banana Pro as the generator) and 65.66% on ImageReq-Bench, and we compare system outcomes across judge-training configurations.