Multinex: An ultra lightweight AI model advancing low light image enhancement

2026-06-08 · Source: News on Artificial Intelligence and Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, short

Summary

Multinex, an ultra-lightweight AI model developed by a University of Manchester student, significantly advances low-light image enhancement (LLIE) by transforming dark, noisy footage into clear, detailed images. Presented at CVPR 2026 (June 3–7), Multinex outperforms comparable compact systems like PairLIE (330K parameters) and ZeroDCE (80K parameters) while using substantially fewer parameters. It is released in a lightweight version (45K parameters) and an extremely compact nano version (0.7K parameters). The model's design leverages classical color vision theory and the Retinex framework, prioritizing enhancement over reconstruction to achieve strong illumination correction and color fidelity. This approach makes it suitable for real-time AI in safety-critical visual systems, despite facing challenges with severe spectral distortions or mixed lighting.

Key takeaway

For Machine Learning Engineers developing real-time visual AI systems, Multinex offers a compelling solution for low-light image enhancement. You should consider its lightweight (45K parameters) or nano (0.7K parameters) versions to achieve significant performance improvements and reduce computational load compared to existing models. This approach allows you to deploy robust visual AI in energy-efficient, safety-critical applications, extending capabilities for autonomous operation in challenging lighting conditions.

Key insights

Multinex uses classical color vision and Retinex principles for efficient, compact low-light image enhancement.

Principles

• Better problem formulation leads to efficient models.
• Integrate classical knowledge into modern AI systems.
• Prioritize enhancement over reconstruction.

Method

Multinex decomposes images into illumination and reflectance using the Retinex framework, applying lightweight neural operations to focus on enhancement.

In practice

• Enhance images for photography and security applications.
• Deploy real-time AI in safety-critical visual systems.
• Extend framework to intrinsic image decomposition or haze removal.

Topics

• Low-Light Image Enhancement
• Multinex
• Retinex Framework
• Lightweight AI Models
• Computer Vision
• Real-time AI

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• DeepGen 1.0: A Lightweight Unified Multimodal Model for Advancing Image Generation and Editing — DeepGen 1.0 is a lightweight 5B multimodal model outperforming larger counterparts in image generation and editing.
• PIXLRelight: Controllable Relighting via Intrinsic Conditioning — PIXLRelight enables physically accurate, controllable single-image relighting by bridging PBR and learned synthesis via intrinsic conditioning.
• SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning - AI at Meta — SAM 3.1 enhances real-time video object tracking via multiplexing, doubling processing speed and reducing GPU demands.
• Unlocking Multi-Spectral Data for Multi-Modal Models with Guided Inputs and Chain-of-Thought Reasoning — Generalist LMMs can interpret multi-spectral data zero-shot by converting it to pseudo-images and using detailed, CoT-enhanced prompts.
• DUALVISION: RGB-Infrared Multimodal Large Language Models for Robust Visual Reasoning — Integrating infrared and RGB data via localized cross-attention significantly boosts MLLM robustness in degraded visual conditions.
• Bridging the Training-Deployment Gap: Gated Encoding and Multi-Scale Refinement for Efficient Quantization-Aware Image Enhancement — QAT and hierarchical architecture improve mobile image enhancement quality and efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by News on Artificial Intelligence and Machine Learning.