SCOPE: Structured Decomposition and Conditional Skill Orchestration for Complex Image Generation

2026-05-08

AI Summary

This work addresses the challenge of maintaining semantic fidelity throughout complex image generation, where inconsistent semantic commitments often lead to distorted visual intent. The authors propose SCOPE, a novel framework that formally characterizes the “concept gap” problem and introduces structured specification modeling to encode semantic commitments. To mitigate commitment drift, SCOPE incorporates a conditional skill orchestration mechanism that dynamically invokes retrieval, reasoning, and repair modules as needed. This design enables end-to-end traceability and consistent enforcement of semantic commitments. Experimental results demonstrate that SCOPE achieves an EGIP score of 0.60 on the Gen-Arena benchmark, substantially outperforming baseline methods, and also excels on WISE-V (0.907) and MindBench (0.61), highlighting its effectiveness in preserving semantic coherence across diverse evaluation settings.

Abstract

While text-to-image models have made strong progress in visual fidelity, faithfully realizing complex visual intents remains challenging because many requirements must be tracked across grounding, generation, and verification. We refer to these requirements as semantic commitments and formalize their lifecycle discontinuity as the Conceptual Rift, where commitments may be locally resolved or checked but fail to remain identifiable as the same operational units throughout the generation lifecycle. To address this, we propose SCOPE, a specification-guided skill orchestration framework that maintains semantic commitments in an evolving structured specification and conditionally invokes retrieval, reasoning, and repair skills around unresolved or violated commitments. To evaluate commitment-level intent realization, we introduce Gen-Arena, a human-annotated benchmark with entity- and constraint-level specifications, together with Entity-Gated Intent Pass Rate (EGIP), a strict entity-first pass criterion. SCOPE substantially outperforms all evaluated baselines on Gen-Arena, achieving 0.60 EGIP, and further achieves strong results on WISE-V (0.907) and MindBench (0.61), demonstrating the effectiveness of persistent commitment tracking for complex image generation.

Problem

• semantic commitments
• Conceptual Rift
• complex image generation
• intent realization
• structured specification

Innovation

• structured decomposition
• conditional skill orchestration
• semantic commitments
• conceptual rift
• specification-guided generation