Portrait composition plays a central role in portrait aesthetics and visual communication, yet existing datasets and benchmarks mainly focus on coarse aesthetic scoring, generic image aesthetics, or unconstrained portrait generation. This limits systematic research on structured portrait composition analysis and controllable portrait generation under explicit composition requirements. In this paper, we introduce PortraitCraft, a unified benchmark for portrait composition understanding and generation. PortraitCraft is built on a dataset of approximately 50,000 curated real portrait images with structured multi-level supervision, including global composition scores, annotations over 13 composition attributes, attribute-level explanation texts, visual question answering pairs, and composition-oriented textual descriptions for generation. Based on this dataset, we establish two complementary benchmark tasks for composition understanding and composition-aware generation within a unified framework. The first evaluates portrait composition understanding through score prediction, fine-grained attribute reasoning, and image-grounded visual question answering, while the second evaluates portrait generation from structured composition descriptions under explicit composition constraints. We further define standardized evaluation protocols and provide reference baseline results with representative multimodal models. PortraitCraft provides a comprehensive benchmark for future research on fine-grained portrait understanding, interpretable aesthetic assessment, and controllable portrait generation.
Figures & tables
Figure 1: Overview of the PortraitCraft benchmark. PortraitCraft is built on 50,000 curated real portrait images and provides a unified framework for portrait composition understanding and generation. The benchmark includes two related tasks. Track 1 evaluates portrait composition understanding through overall score prediction, fine-grained attribute-level analysis, and image-based visual question answering. Track 2 evaluates portrait composition generation from structured composition descriptions under explicit composition constraints. Together, these two tasks connect composition analysis with composition-guided generation and support structured research on portrait composition.
Dataset
Domain
Scale
Expert
Annotation
Attributes
Understanding
Generation
AVA [ 5 ]
Generic Aesthetics
255,528
✗
Overall Score
None
Limited
✗
Flickr-AES [ 17 ]
Generic Aesthetics
40,499
✗
Overall Score
None
Limited
✗
AADB [ 11 ]
Generic Aesthetics
10,000
✗
Overall Score, Attributes
Limited
✓
✗
ArtiMuse [ 2 ]
Generic Aesthetics
10,000
✓
Overall Score, Attributes
8 Attributes
✓
✗
APDDv2 [ 10 ]
Artistic Images
10,023
✓
Overall Score, Attributes, Comments
10 Attributes
✓
✗
PIQ23 [ 3 ]
Portrait Quality
5,116
✓
Quality Labels
3 Attributes
Limited
✗
Table 1: Comparison of PortraitCraft with representative aesthetic and composition-related datasets. Understanding indicates whether the dataset supports structured composition analysis beyond coarse overall scoring.
Figure 2: Statistics of Track 1 composition annotations. (a) Smoothed density distributions of global composition scores for the training and test sets. The test set shows a broader distribution and slightly lower average scores, indicating increased evaluation difficulty. (b) Radar plot of mean scores across the 13 composition attributes. The variation across attributes reflects the heterogeneous characteristics of portrait composition and supports fine-grained evaluation beyond a single overall score.
Type
Samples
Mean
Std
Median
Range
Avg. Attr. Length
Train
4,500
2237.0
251.4
2214
1459–3389
172.0
Test
500
2230.8
240.2
2215
1567–3163
171.2
Overall
5,000
2236.4
250.3
2214
1459–3389
171.9
Table 2: Statistics of structured composition descriptions in PortraitCraft for Track 2. Mean, Std, Median, and Range summarize the distribution of total description length for each dataset split. Avg. Attr. Length denotes the average description length per composition attribute across the 13 predefined attributes. Lengths are measured in characters.
Method
SRCC ↑
PLCC ↑
Level Acc ↑
QA Acc ↑
Qwen3-VL-4B (Zero-shot)
0.6726
0.6420
0.4290
0.7825
Qwen3-VL-4B (Fine-tuned)
0.7778
0.7477
0.4592
0.8765
Table 3: Baseline results on Track 1: Portrait Composition Understanding. All metrics are higher-is-better. Lightweight fine-tuning MLLMs on PortraitCraft improves performance over the zero-shot setting.
Figure 3: Qualitative results on Track 2: Portrait Composition Generation. Two representative examples are shown. For each example, Original denotes the reference portrait image associated with the structured composition description, and Generated denotes the image produced by the model from the same composition specification. The comparison shows that the generated portraits follow key composition cues such as subject placement, spatial organization, and visual emphasis, demonstrating the feasibility of composition-aware portrait generation under the proposed benchmark setting.
Compositional image generation studies how to combine objects, attributes, spatial relations, and visual references into coherent, controllable images.
This paper presents an overview of the inaugural PortraitCraft Challenge, held as one of the official competitions at CVPR 2026. The challenge focuses on portrait composition understanding and generation, aiming to advance AI research in portrait aesthetics analysis and controllable image synthesis. Unlike existing datasets and tasks that primarily focus on global aesthetic scoring, PortraitCraft introduces a unified evaluation framework comprising two complementary tracks. Track 1 requires models to perform structured portrait composition understanding, and Track 2 requires models to generate portrait images from structured composition descriptions under explicit compositional constraints. To support the challenge, we constructed and publicly released a large-scale portrait composition dataset consisting of approximately 50,000 curated real portrait images, providing multi-level supervision. This report describes the challenge setup, evaluation protocols, dataset composition, and final results, along with an analysis of the technical characteristics of the submitted solutions. The PortraitCraft Challenge provides a standardized and reproducible platform for research on portrait composition understanding and generation, and is expected to foster further progress in the fields of portrait aesthetics and controllable image generation.
Portrait aesthetic assessment assigns comparable scores according to how effectively human-centered images fulfill their photographic intent. These scores support data filtering, candidate selection, and preference modeling in image-generation pipelines. Existing methods typically predict a single aesthetic score or use general-purpose MLLMs without conditioning on photographic intent. This omission matters because the same blur, pose, lighting, or framing choice may serve one photographic intent but undermine another. These models thus learn context-agnostic aesthetic priors and yield inconsistent, inaccurate, misleading judgments for portraits with distinct photographic objectives. We introduce PortraitAes-Bench, an 11K-scale benchmark that decomposes this task into intent-conditioned subjudgments. Expert-authored rubrics define nine photographic intents, six first-level dimensions, and 22 secondary criteria. They support a structured pipeline for intent routing, specialist assessment, verification, and score fusion. Following this structure, we train PortraitAes with multi-task supervision. We then improve score comparability through Gaussian score calibration and within-dimension cross-image ranking. On the standard benchmark, PortraitAes achieves a Pearson correlation of 0.924 and a Spearman rank correlation of 0.934. On the hard-case set, its Pearson correlation is 0.829 and its Spearman rank correlation is 0.795. Across both sets, PortraitAes outperforms the evaluated general-purpose MLLMs and specialized aesthetic baselines.
Junzhou Xie, Haozhong Xiong, Xunyun Tian +8
Southeast University · Qwen Business Unit of Alibaba
Text-to-image diffusion models enable personalization of specific visual concepts from a small number of reference images. However, generating a single image that contains multiple personalized subjects, each bound to user-specified attributes such as clothing, accessories, and held objects, remains largely unaddressed. Without explicit spatial constraints, concurrently activated concept checkpoints produce overlapping cross-attention responses, causing per-subject identity degradation and attribute misalignment. Moreover, no established benchmark jointly evaluates these two failure modes in the personalized multi-subject setting. We present MultiCompose, a composition framework that decouples per-concept personalization from multi-subject inference. A semantic preservation regularization maintains attribute binding capacity during fine-tuning, while a two-phase inference procedure automatically establishes subject layout and composes per-concept predictions through spatially exclusive masks. We further introduce MSP-Bench, a benchmark that jointly evaluates identity fidelity (ID), attribute binding accuracy (BIND), and attribute misalignment (MIS) through a dual-pathway protocol. Experiments show that MultiCompose outperforms existing methods on both conventional metrics and MSP-Bench, confirming the benchmark's ability to reveal failure modes that conventional metrics overlook. Code is available at https://github.com/I2-Multimedia-Lab/MultiCompose
Ruirui Zhang, Zhengkai Zhao, Pan Gao
Nanjing University of Aeronautics and Astronautics College of Computer Science and Technology Nanjing, Jiangsu, China · Nanjing University of Aeronautics and Astronautics College of Artificial Intelligence Nanjing, Jiangsu, China