Scone: Bridging Composition and Distinction in Subject-Driven Image Generation via Unified Understanding-Generation Modeling
This addresses a limitation in subject-driven image generation for realistic applications, though it appears incremental as it builds on existing multi-subject composition approaches.
The paper tackles the problem of subject-driven image generation where existing methods handle multi-subject composition but neglect distinction (identifying correct subjects among multiple candidates), limiting effectiveness in complex visual settings. The proposed Scone method outperforms existing open-source models in both composition and distinction tasks on two benchmarks.
Subject-driven image generation has advanced from single- to multi-subject composition, while neglecting distinction, the ability to identify and generate the correct subject when inputs contain multiple candidates. This limitation restricts effectiveness in complex, realistic visual settings. We propose Scone, a unified understanding-generation method that integrates composition and distinction. Scone enables the understanding expert to act as a semantic bridge, conveying semantic information and guiding the generation expert to preserve subject identity while minimizing interference. A two-stage training scheme first learns composition, then enhances distinction through semantic alignment and attention-based masking. We also introduce SconeEval, a benchmark for evaluating both composition and distinction across diverse scenarios. Experiments demonstrate that Scone outperforms existing open-source models in composition and distinction tasks on two benchmarks. Our model, benchmark, and training data are available at: https://github.com/Ryann-Ran/Scone.