Researchers introduce StruPO, a structure-aware prompt editor for text-to-image models
A new arXiv paper proposes StruPO, a structure‑aware prompt editor that leverages a calibrated scene graph to guide text‑to‑image generation. The graph identifies elements that must stay unchanged, while token‑level deletions, reorderings, and insertions modify the prompt without altering the requested scene. StruPO turns external structural feedback into a training signal for a reusable editing…
Key points
- StruPO uses a calibrated scene graph to condition prompt edits while preserving scene structure.
- Experiments on three 2K‑prompt test sets and four frozen backbones show consistent gains in relational fidelity.
- Human evaluation with 15 annotators reports higher preference for StruPO‑edited prompts, especially on complex relations.
The authors evaluate StruPO on three disjoint test sets of 2,000 prompts each and four frozen generation backbones, comparing it against learned prompt optimizers, iterative structural feedback, diffusion‑model fine‑tuning, and attention‑control methods. Results show consistent improvements in the faithfulness‑aesthetics trade‑off and higher relational fidelity. Independent automatic metrics and a study with 15 annotators confirm the gains, with preference margins growing as relation complexity increases. Ablation experiments link the improvements to graph conditioning, local editing, and image‑grounded learning, demonstrating that a single‑pass, graph‑calibrated edit can replace iterative critique‑rewrite loops.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Anthropic says unreleased model built working exploits for Chrome, Safari, Windows, Linux · 2 src
- OpenAI and Anthropic claim their AI models refuted the Exponential Time Hypothesis · 13 src
- Cprisp framework predicts circRNA-RBP interactions for neurodegeneration research · 1 src
- TinyLeague updates benchmarks: drops AA Index, adds KV efficiency metrics · 1 src
- Artificial Analysis releases Intelligence Index v5 with private datasets · 1 src
Comments
via GitHub Discussions