Training Free Stylized Abstraction

TL;DR

This paper introduces a training-free method for stylized abstraction that generates visually exaggerated yet semantically faithful images from a single input, leveraging vision-language models and a novel flow inversion strategy.

Contribution

It presents a novel inference-time framework for stylized abstraction that does not require training, using style-aware features and a new rectified flow inversion for structure reconstruction.

Findings

Effective across diverse styles like LEGO and South Park

Generalizes well to unseen identities and styles

Supports multi-round abstraction without fine-tuning

Abstract

Stylized abstraction synthesizes visually exaggerated yet semantically faithful representations of subjects, balancing recognizability with perceptual distortion. Unlike image-to-image translation, which prioritizes structural fidelity, stylized abstraction demands selective retention of identity cues while embracing stylistic divergence, especially challenging for out-of-distribution individuals. We propose a training-free framework that generates stylized abstractions from a single image using inference-time scaling in vision-language models (VLLMs) to extract identity-relevant features, and a novel cross-domain rectified flow inversion strategy that reconstructs structure based on style-dependent priors. Our method adapts structural restoration dynamically through style-aware temporal scheduling, enabling high-fidelity reconstructions that honor both subject and style. It supports multi-round abstraction-aware generation without fine-tuning. To evaluate this task, we introduce StyleBench, a GPT-based human-aligned metric suited for abstract styles where pixel-level similarity fails. Experiments across diverse abstraction (e.g., LEGO, knitted dolls, South Park) show strong generalization to unseen identities and styles in a fully open-source setup.