Aiarty Video Enhancer

Aiarty splits its work across three models rather than one catch-all. moDetail-HQ v3 leans on Diffusion plus GAN to add sharpness and texture, and suits hair, skin and outdoor detail. Smooth-HQ v3 aims for faithful restoration with natural deblurring and color realism. superVideo vHQ is the hybrid model built for the hardest cases, the grainy low-light and dark interior clips where other tools give up. Picking the right one is the single biggest lever on output quality.

The software gives you a speed-versus-quality dial with two paths. Turbo optimizes the AI inference for speed and works well as a default, while Step mode runs tiled inference that refines quality further and cuts VRAM use, which matters for long or high-resolution jobs. The trade is straightforward. Step is slower, Turbo gets you through a batch faster.

The interpolation module generates new frames between existing ones to smooth motion and support slow motion. It targets stuttering footage and can push output to 90fps, 120fps or higher, using flow estimation and motion compensation so the added frames don't smear or distort. It fits fast action, timelapse and animation as much as live footage.

Alongside the picture work, the software analyzes audio tracks and removes background noise while lifting vocals. It's built for the everyday cases where a good image is paired with hiss, wind or room noise, such as vlogs, podcasts, lectures and family recordings. It's a one-click pass rather than a full audio editor.

A color panel covers white balance, exposure, contrast, saturation, highlights and shadows, plus rotate, flip, crop and deinterlace. The SDR-to-HDR feature converts 8-bit footage to 10-bit, with three modes, Neutral, Bright and Graded, and a strength slider to keep the effect from tipping into over-processing. It's a quick grade, not a colorist's toolkit, but it covers the common fixes.

You can queue hundreds of clips in one session, set the model and output folder once, and export the whole list. Everything runs locally on your own hardware, so footage never gets uploaded to a server. That combination, bulk handling without an internet round trip, is the core reason people pick a desktop tool over a browser-based upscaler.

The engine is built to lean on your graphics card, reaching up to 95% GPU utilization. On Windows it supports DirectML and TensorRT, and on Mac it uses Metal and the Neural Engine with native CoreML support. NVIDIA RTX 30, 40 and 50 series are called out as the fastest path, while AMD and Intel cards are supported through DirectML. A newer GPU is the main thing that decides how quick a render feels.