LimitlessLab Unveils AI Upscaling Models for True-to-Life Detail

    
        By vramkickedin    
     | 
    
            June 22, 2026 at 9:27 pm        
    
     | 
    
        2 min read

A new release of AI Upscaling Models from Limitlesslab provides a set of custom-trained neural networks for photorealistic image and video enhancement. The collection includes GAN models built on architectures like ESRGAN, SAFMN, and RealPLKSR, each designed to handle different types of degradation. They upscale images while preserving fine textures such as skin, fabric, and natural grit.

The developer trained all models on a personally curated library of over 11,750 high-quality real-world HD images. This effort, spanning several months, focused on creating versatile models that intelligently remove compression artifacts, noise, and blur. The tools can be used directly within popular interfaces like Chainner or ComfyUI.

Models trained on hand-picked HD dataset

Key Features

Photorealistic upscaling with authentic texture rendering.
Removes JPEG, WEBP, AVC, HEVC compression artifacts.
Handles noise and grain without losing detail.
Trained on 11,750+ hand-selected HD images.
Works with Chainner and ComfyUI workflows.
ONNX models available for TensorRT acceleration.
Deblurring and detail restoration for degraded media.

Anyone needing to upscale photos, game textures, or video frames can benefit from these models. They are especially useful for restoring older compressed images or preparing footage for higher resolutions. The ONNX versions further allow real-time processing on supported NVIDIA hardware.

Developer's vision and software plans

The creator is also building a standalone high-end video restoration and upscaling application, taking aim at commercial tools. To use the PureScale 2.0 models in Chainner, you must replace the default SAFMN architecture file with the provided safmn.py. Future updates will target even higher fidelity and better generalization across diverse image types.