aiqbench analysis
Scone addresses a critical gap in subject-driven image generation by unifying multi-subject composition with the ability to distinguish specific candidates, a capability often overlooked in current models. The approach utilizes a two-stage training scheme where an understanding expert acts as a semantic bridge to guide generation while minimizing interference between subjects. Alongside the model, the authors release SconeEval, a new benchmark designed to rigorously evaluate these dual capabilities across diverse visual scenarios.
Scone: Bridging Composition and Distinction in Subject-Driven Image Generation via Unified Understanding-Generation Modeling
arXiv:2512.12675v3 Announce Type: replace-cross Abstract: Subject-driven image generation has advanced from single- to multi-subject composition, while neglecting distinction, the ability to distinguish 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.
Abstract:Subject-driven image generation has advanced from single- to multi-subject composition, while neglecting distinction, the ability to distinguish 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: this https URL.
CVPR 2026 Highlight. Code: this https URL
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2512.12675 [cs.CV]
(or arXiv:2512.12675v3 [cs.CV] for this version)
Full text under the publisher's open licence · 157 words