WorldSR: Harnessing World Knowledge Search for Grounded Image Super-Resolution

Abstract

Generative-prior-based super-resolution (SR) methods leverage implicit knowledge learned from large-scale models to enhance low-resolution inputs. However, such priors are inherently stochastic and optimized for diverse generation, conflicting with the deterministic, high-fidelity reconstruction required for SR. It is also tied to the base model, costly to update, and prone to hallucinations as real-world data evolves.

To address these limitations, we introduce WorldSR, a novel framework that augments SR with explicit, entity-aligned visual world knowledge. WorldSR automatically decomposes the input into semantic entities and performs entity-wise retrieval of structurally and semantically aligned images from a large-scale external corpus. For efficiency and reliability, we incorporate a memory mechanism to reuse previously retrieved results and a multi-step filtering process to refine candidate references. The selected references as world knowledge are then integrated through a novel multi-reference super-resolution model, which provides fine-grained constraints for detail synthesis and structural recovery. Experiments on real-world SR benchmarks show that WorldSR consistently improves structural fidelity and visual quality, outperforming existing generative super-resolution methods. Our results suggest that entity-level grounding in external visual world knowledge provides a simple and effective complement to model-bound generative priors.

Quantitative Results

Comparison with state-of-the-art methods on RealLQ250, Bird, Face, WRSR, and CUFED5. WorldSR establishes a new state of the art across all no-reference metrics.

Red / blue: best / second-best performance. WorldSR surpasses the second-best face-specific approach by +6.4% IDS on the Face dataset, and is the best performer across all no-reference metrics on CUFED5.

Applications

Beyond benchmarks, WorldSR plugs into real-world applications: for each degraded input, a sharp reference is retrieved via entity-level search, and the world-grounded SR model restores it. Inside a comparison: scroll to zoom, drag to pan, drag the round handle to wipe. Click any other image to enlarge.

Smartphone 100× Telephoto Restoration

A 100× telephoto shot taken by a smartphone is severely blurred. WorldSR restores it with a DSLR-quality photo of the same poster retrieved by entity-level search. Drag the slider to compare.

Phone 100× Telephoto WorldSR

DSLR Photo

Phone Telephoto WorldSR

Reference

Example 1 / 2

ID Photo Enhancement

WorldSR fixes the identity-fidelity (ID) issue of Nano Banana-generated ID photos, restoring the person's true identity with a web-retrieved reference of the same person. Drag the slider to compare.

Nano Banana WorldSR

Reference

Nano Banana WorldSR

Reference

Example 1 / 2

AIGC Repair

A degraded AI-generated artwork is repaired and re-rendered at a higher resolution, using its clear original as the reference. Drag the slider to compare.

AIGC Gen WorldSR

Reference

3D Gaussian Splatting Refinement

Blurry, floaty 3DGS renders are cleaned up and sharpened with a sharp reference frame. Drag the slider to compare.

3DGS Render WorldSR

Reference