OPRO: Orthogonal Panel-Relative Operators for Panel-Aware In-Context Image Generation

TL;DR

OPRO introduces a parameter-efficient method for panel-aware in-context image generation, using orthogonal operators on frozen positional encodings to improve editing pipelines without retraining the entire model.

Contribution

The paper proposes a novel orthogonal operator-based adaptation for diffusion transformers that enhances panel-relative conditioning in image generation tasks.

Findings

Method generalizes across various positional encoding regimes.

Improves in-context image editing pipelines.

Maintains pre-trained intra-panel synthesis behavior.

Abstract

We introduce a parameter-efficient adaptation method for panel-aware in-context image generation with pre-trained diffusion transformers. The key idea is to compose learnable, panel-specific orthogonal operators onto the backbone's frozen positional encodings. This design provides two desirable properties: (1) isometry, which preserves the geometry of internal features, and (2) same-panel invariance, which maintains the model's pre-trained intra-panel synthesis behavior. Through controlled experiments, we demonstrate that the effectiveness of our adaptation method is not tied to a specific positional encoding design but generalizes across diverse positional encoding regimes. By enabling effective panel-relative conditioning, the proposed method consistently improves in-context image-based instructional editing pipelines, including state-of-the-art approaches.