IRIS by Wavelet Beam
IRIS Video Denoising: Remove Noise, Keep Every Detail
IRIS is GPU-accelerated video denoising for broadcast, streaming, post-production and archives. It separates noise from real image detail, so you get a cleaner picture, a higher perceived resolution and up to 30% lower bitrates, fully automated on NVIDIA GPUs: on-premises, in the cloud or as SaaS.
IRIS at a glance
- Video denoising without blurring: noise and grain are removed, fine textures and edges stay.
- Up to 30% lower bitrate at the same or better perceived quality, for H.264, HEVC, AV1 and VVC.
- Fully automatic: the noise profile is selected per shot, no manual tuning.
- Resolution independent up to 8K, SDR and HDR, on NVIDIA GPUs with CUDA.
- Runs anywhere: on-premises, cloud, hybrid or SaaS with per-minute pricing, via Kubernetes, Docker or VMs.
What is video denoising?
Video denoising removes unwanted noise from moving pictures: sensor noise from cameras, gain noise in low light, film grain from scans and the noise that previous compression steps leave behind. Spatial methods clean each frame on its own, temporal methods compare neighbouring frames, and spatio-temporal methods combine both. The goal is always the same: keep the real picture and drop the random part that carries no information.
Why classic denoisers are not enough
Image information is part of the video noise floor. Fine textures, skin detail and film structure sit at the same amplitude as the noise. A classic filter cannot tell them apart and smooths both: the picture looks cleaner, but it also looks softer, and the resolution you paid for at the camera is gone. Even compression ratios below 10:1 cost image information, which is why "visually lossless" is a misleading term.
IRIS takes a different approach. It does not just reduce visible noise, it lifts the signal out of the noise floor. Image information that would otherwise be lost or unusable is recovered, and the perceived resolution goes up instead of down.
What video denoising with IRIS gives you
Higher perceived resolution
Clean edges and preserved textures look sharper on every screen. A signal produced in 4K can be denoised once and downscaled for every lower resolution, so second screens without their own denoising or deblocking get the best possible version too.
Lower bitrates
Noise is random and cannot be predicted from previous frames, so an encoder spends bits on it in every frame. Noise is also mistaken for motion, which leads to faulty motion vectors. Block-based codecs suffer most: noise amplifies the distortions at block boundaries. Removing noise first frees the encoder for real image content and enables up to 30% lower bitrates. In a statistical multiplex, one noisy service no longer takes bandwidth away from the others. The details are in our guide why denoise video before encoding.
Fewer quality drops
In a critical low-light scene of Netflix "Meridian" at 9.6 Mbit/s in UHD HEVC, IRIS reduced the share of frames below VMAF 93 from 25.73% to 13.29% and the share of frames below 40 dB PSNR from 51.62% to 10.15%. Combined with per-shot encoding, the same content reached UHD at 7.54 Mbit/s with a VMAF of 96.02.
Better HDR and cleaner AI input
HDR shows more detail in dark areas, where noise amplitudes are high compared to the image. High-fidelity denoising without blurring gives colorists and compositing artists the most room to work. The same clean signal helps AI models: they learn scene content instead of sensor noise, see IRIS for AI.
Where to use it in your workflow
IRIS does not need to run at every stage. You choose the most effective entry point:
- Source driven: denoise RAW data before debayering with IRIS.RAW and BRAW2IRIS, for the best signal-to-noise ratio in grading.
- Master driven: clean an intermediate such as FFV1 or ProRes after grading and VFX.
- Distribution driven: denoise per shot right before encoding, with IRIS.ANALYST selecting the noise profile.
- Live: real-time video denoising in the signal path with IRIS.Broadcast, also as a native MXL Media Function.
- Archives: homogenize a VOD catalog and restore film scans, massive archives can be processed on a single system.
How IRIS runs
IRIS processes any FFmpeg-readable format, including FFV1 and ProRes intermediates, resolution independent up to 8K. It is built on thousands of lines of CUDA code and scales across multiple GPUs. Deployment is fully automated for clusters, cloud and on-premises with Kubernetes, Docker or VMs. Our SaaS offering provides video denoising workflows at a competitive per-minute rate.
Test IRIS on your own material
Send us a sample and we compare a standard encode with IRIS video denoising at the same settings, measured with VMAF, PSNR and QBVE. Contact info@waveletbeam.com.