CREward: A Type-Specific Creativity
Reward Model

Jiyeon Han1, Ali Mahdavi-Amiri1, Hao Zhang1, Haedong Jeong2
1Simon Fraser University 2Sogang University
CVPR 2026
Paper Code arXiv

🤗

 Dataset
First research result visualization

We propose to assess and guide creative image generation along three visual dimensions. We show strong alignment between human and LVLM preferences on type-specific creativity and use LVLMs to build a corresponding preference dataset. We then train CREward, a type-specific reward model. CREward enables evaluation and comparison of generative models, filtering of highly creative samples for design, and guidance of type-specific generation (e.g., via LoRA). It can also be combined with XAI methods to support explainable creativity.

Abstract

Creativity is a complex phenomenon. When it comes to representing and assessing creativity, treating it as a single undifferentiated quantity would appear naive and underwhelming. In this work, we learn the first type-specific creativity reward model, coined CREward, which spans three creativity "axes," geometry, material, and texture, to allow us to view creativity through the lens of the image formation pipeline. To build our reward model, we first conduct a human benchmark evaluation to capture human perception of creativity for each type across various creative images. We then analyze the correlation between human judgments and predictions by large vision-language models (LVLMs), confirming that LVLMs exhibit strong alignment with human perception. Building on this observation, we collect LVLM-generated labels to train our CREward model that is applicable to both evaluation and generation of creative images. We explore three applications of CREward: creativity assessment, explainable creativity, and creative sample acquisition for both human design inspiration and guiding creative generation through low-rank adaptation.

CREward: Type-specific Creativity Reward Model

First research result visualization

We train a type-specific creativity reward model, CREward, by leveraging T2I models and LVLMs to obtain large-scale preference labels on synthetic, type-specific creative images.

Type-specific Creativity Assessment

Human-AI Co-Creation

LoRA Sliders for Type-specific Creative Generation

First research result visualization

CREward can guide type-specific generation via LoRA sliders, enabling single or combined types at user-specified strengths.

Explainable Creativity

CREward can be extended, with XAI methods, to identify and analyze which regions of an image contribute to creativity for a target type.

BibTeX

@misc{han2025creward,
      title={CREward: A Type-Specific Creativity Reward Model}, 
      author={Jiyeon Han and Ali Mahdavi-Amiri and Hao Zhang and Haedong Jeong},
      year={2025},
      eprint={2511.19995},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.19995}, 
}