# TRACE-Bench: Decomposing and Diagnosing Multi-Reference Image Generation

arXiv: https://arxiv.org/abs/2608.16765 · HTML: https://mlsanity.com/paper/2608.16765/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 30 of its 59 references that are in the map.
- First posted on arXiv 17 Aug 2026.

## Builds on (most cited first, 30 of 30)

- [Hierarchical Text-Conditional Image Generation with CLIP Latents](https://mlsanity.com/paper/2204.06125/) (2022)
- [Scaling Rectified Flow Transformers for High-Resolution Image Synthesis](https://mlsanity.com/paper/2403.03206/) (2024)
- [DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation](https://mlsanity.com/paper/2208.12242/) (2022)
- [An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion](https://mlsanity.com/paper/2208.01618/) (2022)
- [VBench: Comprehensive Benchmark Suite for Video Generative Models](https://mlsanity.com/paper/2311.17982/) (2023)
- [FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space](https://mlsanity.com/paper/2506.15742/) (2025)
- [Emerging Properties in Unified Multimodal Pretraining](https://mlsanity.com/paper/2505.14683/) (2025)
- [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://mlsanity.com/paper/2202.11094/) (2022)
- [ELLA: Equip Diffusion Models with LLM for Enhanced Semantic Alignment](https://mlsanity.com/paper/2403.05135/) (2024)
- [FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention](https://mlsanity.com/paper/2305.10431/) (2023)
- [VIEScore: Towards Explainable Metrics for Conditional Image Synthesis Evaluation](https://mlsanity.com/paper/2312.14867/) (2023)
- [Emu3.5: Native Multimodal Models are World Learners](https://mlsanity.com/paper/2510.26583/) (2025)
- [KRIS-Bench: Benchmarking Next-Level Intelligent Image Editing Models](https://mlsanity.com/paper/2505.16707/) (2025)
- [DiffusionSeg: Adapting Diffusion Towards Unsupervised Object Discovery](https://mlsanity.com/paper/2303.09813/) (2023)
- [ReMamber: Referring Image Segmentation with Mamba Twister](https://mlsanity.com/paper/2403.17839/) (2024)
- [FireRed-Image-Edit-1.0 Technical Report](https://mlsanity.com/paper/2602.13344/) (2026)
- [Easier Painting Than Thinking: Can Text-to-Image Models Set the Stage, but Not Direct the Play?](https://mlsanity.com/paper/2509.03516/) (2025)
- [AttrSeg: Open-Vocabulary Semantic Segmentation via Attribute Decomposition-Aggregation](https://mlsanity.com/paper/2309.00096/) (2023)
- [Multi-Modal Prototypes for Open-World Semantic Segmentation](https://mlsanity.com/paper/2307.02003/) (2023)
- [DeepGen 1.0: A Lightweight Unified Multimodal Model for Advancing Image Generation and Editing](https://mlsanity.com/paper/2602.12205/) (2026)
- [GEditBench v2: A Human-Aligned Benchmark for General Image Editing](https://mlsanity.com/paper/2603.28547/) (2026)
- [Boundary-aware Supervoxel-level Iteratively Refined Interactive 3D Image Segmentation with Multi-agent Reinforcement Learning](https://mlsanity.com/paper/2303.10692/) (2023)
- [UniReason 1.0: A Unified Reasoning Framework for World Knowledge Aligned Image Generation and Editing](https://mlsanity.com/paper/2602.02437/) (2026)
- [MICON-Bench: Benchmarking and Enhancing Multi-Image Context Image Generation in Unified Multimodal Models](https://mlsanity.com/paper/2602.19497/) (2026)
- [Contrast-Unity for Partially-Supervised Temporal Sentence Grounding](https://mlsanity.com/paper/2502.12917/) (2025)
- [MoMa: Modulating Mamba for Adapting Image Foundation Models to Video Recognition](https://mlsanity.com/paper/2506.23283/) (2025)
- [MACRO: Advancing Multi-Reference Image Generation with Structured Long-Context Data](https://mlsanity.com/paper/2603.25319/) (2026)
- [H2R-Bench: Benchmarking Human-to-Robot Manipulation Video Generation in World Models](https://mlsanity.com/paper/2608.13049/) (2026)
- [GenMask: Adapting DiT for Segmentation via Direct Mask Generation](https://mlsanity.com/paper/2603.23906/) (2026)
- [Reason, Then Re-reason: Cross-view Revisiting Improves Spatial Reasoning](https://mlsanity.com/paper/2606.11683/) (2026)

## Abstract (arXiv)

Despite recent advances in unified multimodal models for multi-reference image generation, existing benchmarks remain organized around predefined task types (e.g., "subject composition"), which are ill-suited to this combinatorial setting and lead to fragmented coverage, uncontrolled complexity, and little diagnostic value. Recognizing that diverse multi-reference tasks share a common set of atomic operations, we adopt a capability-oriented perspective and formalize four operators: Anchor (f), Disentangle (g), Apply (oplus), and Compose (C). Any multi-reference prompt can then be represented as a compositional formula over these operators, whose structural complexity is quantified by the number of operator slots. Building on this formulation, we construct TRACE-Bench, comprising approximately 1,600 evaluation cases across slot counts 1--8, built from 631 formula templates and around 4,000 reference images spanning diverse artistic styles and real-world subjects. The formula structure directly drives an operator-aligned evaluation protocol for per-capability scoring and a diagnostic tree analysis for recursive failure localization. Evaluating 9 leading models reveals insights invisible to holistic scoring: the primary bottleneck lies in disentanglement (g) and attribute binding (oplus) rather than scene-level composition (C), with even the best model scoring only 0.74 on attribute fidelity. Project page: https://amuseum-whr.github.io/TraceBench