Search

New image analysis method breaks through cryo-EM resolution limits | News | Science Japan

A research group introduces PASR, a post-acquisition super resolution method for cryo-EM that recovers high-frequency details beyond the Nyquist limit by subdividing pixels and using inter-frame shifts during motion correction; it works with standard software, improves resolution without new measurements, and aims to boost efficiency and enable real-time processing.

[2611.06042] UniRestorer++: Scalable Universal Image Restoration with Multi-Granularity Degradation Modeling

UniRestorer++ presents a scalable image-restoration framework using multi-granularity degradation modeling, group-specific experts, and route-based specialization to improve performance, especially on out-of-distribution and real-world data, with a new CDD-50 benchmark and code release.

GitHub - BennyDaBall930/ComfyUI-Open-H3-FaceRegion: Keep faces when you regenerate MiniMax H3 videos to 2K: automatic face mask so faces start NVIDIA's Open-H3-Regenerate-2K pass at lower noise. ComfyUI nodes + one-shot prompt-to-2K workflow. · GitHub

A workflow and node package for ComfyUI that preserves faces during 2K upscaling of H3 videos by applying an automatic face region mask. It uses YuNet to detect faces, maps them onto a 2K latent grid, and applies per-region denoise and strength controls to maintain face detail while keeping overall frame consistency. Requires specific dependencies and other ComfyUI components to run.

CLUBIC 🔵 This app uses AI to enhance your photos and videos without sending them over the Internet!

Translated title. Original: CLUBIC 🔵 Cette appli utilise l’IA pour améliorer vos photos et vidéos sans les envoyer sur Internet !

QualityScaler runs locally on Windows with GPU or CPU, delivering AI-based media enhancement without internet access; it supports NVIDIA, AMD, and Intel GPUs, requires 8 GB RAM and a 4 GB GPU, and offers a free version without cloud processing, though results may introduce artificial textures.

18-year-old from Wolfsburg reaches the final of the national AI competition | ndr.de

Translated title. Original: 18-Jähriger aus Wolfsburg erreicht Finale des KI-Bundeswettbewerbs | ndr.de

Mark Leupold, 18, qualified for the national AI competition final with a system that sharpens satellite images and estimates how reliable the added detail is. The jury cited possible uses in agriculture, disaster response, and environmental monitoring; Leupold did not win a prize.

NVIDIA App 11.0.9 adds support for DLSS Override in 19 more games

Translated title. Original: NVIDIA App 11.0.9 añade soporte para DLSS Override en 19 juegos más

NVIDIA App 11.0.9.585 adds 19 new DLSS Override-supported games and six games to the automatic optimization feature, allowing users to use newer DLSS versions without patches and to auto-tune graphics per hardware; the update also includes unspecified stability and security fixes.

18-year-old from Wolfsburg reaches final of national AI competition | tagesschau.de

Translated title. Original: 18-Jähriger aus Wolfsburg erreicht Finale des KI-Bundeswettbewerbs | tagesschau.de

Mark Leupold, an 18-year-old from Wolfsburg, reached the national AI competition final with a system that sharpens satellite images and estimates the accuracy of its results. The jury cited farming, disaster response, and environmental monitoring as possible uses; Leupold received none of the three prizes.

Microsoft’s own upscaler and alternative to FSR just landed on Xbox Ally X handhelds - actually uses an NPU for something useful | PC Gamer

Microsoft has released Auto SR for the ROG Xbox Ally X and X20, adding undocked and battery use, support for up to 120 fps, and Auto HDR. The feature uses an NPU to upscale games, mainly as an option when FSR is unavailable or performs poorly.

Upscale Small or Blurry Images for €0 With Free Upscayl | AIRushX

This guide explains how to install and use the free Upscayl app to enlarge images on a computer, including model choices and output settings. It covers GPU requirements, model licence limits and the risk that AI may invent image details.

Open-H3-Regenerate-2K: High-Resolution Video Refinement with Few-Step Diffusion

Open-H3-Regenerate-2K is a high-resolution video refinement approach that keeps core models frozen, uses a trained refiner LoRA, and upscales latent representations through a fast, three-step denoising process to produce 768P–4K outputs from low-resolution drafts in a single pass. It emphasizes speed versus prior methods and offers an open alternative for the regeneration step starting from an H3 draft.