AI

From Structural Feedback to Prompt Policies: Learning Faithful Text-to-Image Prompt Editors

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 28/10/2026, 04:03:25 UTC.

Briefings

A new method called StruPO uses a scene graph to guide text-to-image prompt editing, improving object fidelity while maintaining global text similarity. The system replaces external critique with a training signal derived from structural feedback, allowing for one-pass optimization after calibration. Experiments tested on three 2K-prompt sets showed gains in relational fidelity compared to baseline methods.

Key points

• StruPO uses a calibrated scene graph as a shared representation for policy conditioning and image-grounded supervision.
• The method improves the faithfulness-aesthetics trade-off with consistent gains in relational fidelity across three 2K-prompt test sets.

Context

This research addresses the tension between improving image fidelity and maintaining global text similarity during prompt editing. The StruPO approach transforms structural feedback from external critique into a training signal, enabling reusable editing policies without iterative critique loops.

Limitations

Evidence is limited to three disjoint 2K-prompt test sets and four frozen generation backbones. Results are based on ablations and automatic evaluators; human preference margins increase with relation complexity but specific margin values are not reported. The study relies on preprint status and does not include independent human preference studies beyond the 15-annotator study.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 28/10/2026, 04:04:18 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeihftastfqlit3i2qlv5rjz6dxrcvxm5rbgizdwzba4x647q7366fe