CatalogStitch

Abstract

Generative object compositing methods have shown remarkable ability to seamlessly insert objects into scenes. However, when applied to real-world catalog image generation, these methods require tedious manual intervention: users must carefully adjust masks when product dimensions differ, and painstakingly restore occluded elements post-generation. We present CatalogStitch, a set of model-agnostic techniques that automate these corrections, enabling user-friendly content creation. Our dimension-aware mask computation algorithm automatically adapts the target region to accommodate products with different dimensions; users simply provide a product image and background, without manual mask adjustments. Our occlusion-aware hybrid restoration method guarantees pixel-perfect preservation of occluding elements, eliminating post-editing workflows. We additionally introduce CatalogStitch-Eval, a 58-example benchmark covering aspect-ratio mismatch and occlusion-heavy catalog scenarios, together with supplementary PDF and HTML viewers. We evaluate our techniques with three state-of-the-art compositing models (ObjectStitch, OmniPaint, and InsertAnything), demonstrating consistent improvements across diverse catalog scenarios. By reducing manual intervention and automating tedious corrections, our approach transforms generative compositing into a practical, human-friendly tool for production catalog workflows.

Dimension-Aware Mask Compositing

Products retain their native proportions with our adapted mask, while freeform and bounding-box masks cause stretching and distortion. Results consistent across all three compositing models.

a

Shoe replacement with different proportions

Background

Product

Freeform

BBox

Ours

InsertAnything

OmniPaint

ObjectStitch

b

Helmet replacement with aspect ratio mismatch

InsertAnything

OmniPaint

ObjectStitch

c

Product swap in lifestyle setting

InsertAnything

OmniPaint

ObjectStitch

Occlusion-Aware Restoration

Foreground occluders are perfectly preserved after compositing via exact pixel restoration. Comparing before and after our restoration step across mask types and compositing models.

a

Sofa with foreground furniture

Background

Product

Occluders

Freeform

BBox

Ours

InsertAnything — Before

InsertAnything — After Restoration

ObjectStitch — Before

ObjectStitch — After Restoration

b

Scene with decorative elements

InsertAnything — Before

InsertAnything — After Restoration

ObjectStitch — Before

ObjectStitch — After Restoration

c

Generated scene with occluding objects

InsertAnything — Before

InsertAnything — After Restoration

ObjectStitch — Before

ObjectStitch — After Restoration

Quantitative Results

Evaluated on CatalogStitch-Eval across five metrics. Our techniques consistently improve every baseline model.

AR Error drops from ~30% to ~4–5% uniformly across all models. InsertAnything + Ours achieves best results on all five metrics.

CatalogStitch-Eval Benchmark

A challenging evaluation benchmark specifically designed for catalog image compositing, covering both core failure modes.

Explore the Benchmark

Citation