We are proud to bring a new Enhancement model into the mix – Nyx – which will focus on denoising for high quality videos. We are also releasing version 2 of Iris, which is a significant upgrade from v1 for most results. Lastly we are adding the capability to correct the rotation of your videos by 90, 180 and 270 degrees. More info about models and frontend change below.
Adds support for rotating videos 90, 180, and 270 degrees.
Adds crash logs to “logs for support” and changes format to zip file.
Adds change moving “Enable Previous Model Versions” preference to advanced section of settings.
Adds change renaming “Second Pass Enhancement” to just “Second Enhancement” to reduce any potential confusion with two-pass encoding.
Fixes bug causing logs to be inaccessible on Windows.
Fixes special characters in file path causing export failure when using Export As.
Fixes crash on app open for some machines.
Fixes trim accuracy issues for ProRes.
Fixes bug causing inaccuracy with video duration after resetting trim range.
Nyx:
A new model for denoising high-quality video.
The focus of the model is on footage from high-quality cameras captured in less-than-ideal lighting conditions.
Processes videos at 1x scale to maintain the level of detail in the input video while removing noise.
Iris v2:
An improved version of Iris v1, offerring more details and sharpness especially to faces.
It uses TensorRT models for Nvidia GPUs and could result in a 50% speed-up.
It also reduces the “previous frame’s ghost” artifacts.
It reduces “marching ant” (diagonal movement of texture) artifacts, especially, for 1x and 2x.
Attention:
The format for support logs is being changed to include license information (auth.tpz). While this file is encrypted and only useful on your machine, please try to avoid posting it or your support logs here by mistake. This would not likely lead to a problem, but better to be safe than sorry. Please submit any logs directly to our support team or devs
Known Issues:
Intel ARC is currently unsupported.
If using Frame Interpolation to increase fps over 2x, output may contain duplicate frames.
For Interlaced and Interlaced-Progressive videos, setting FPS back to Original may cause issues. To get around this, re-import the file. This only happens after manually changing the FPS.
Videos with mismatched metadata and streams will display incorrect duration
Frame number preview length may shorten on app restart
Some vertical videos will be rotated an additional 90 degrees when imported
Just in time to try on these 26 episodes of an anime from the 70s
Just a real quick test before I head to bed… I grabbed a high noise BD clip from Crouching Tiger I have and both iris v2 and nyx do a SUPER impressive job just on auto of fighting noise. Nyx did a better job, absolutely, but iris was sharper.
BOTH of them are about 2x the speed of the previous version too. I am SUPER excited to play with this tomorrow! So far, this looks like a GREAT JOB!
I have a few of those episodes I mentioned up above going for tomorrow, 640x480 to 1920x1080 with Iris v2 on auto and then second pass gia computer generated option. See how they look tomorrow am. Oh and those are cranking at about 8-17fps on my 3080 with my memory set to 65% and ONLY using 8 of my 16 cores (I like to use my pc for work too )
Would the Nyx model be good for removing film grain from 1080p old movies? I find when they transferred movies to blu ray from 90s and before it made the film grain 10x more noticeable than the VHS or DVD versions.
Thanks for the weekly updates, always look forward to Tuesdays. haha
This is the type of situation we expect Nyx to do well with: noisy or grainy sources that still show a high amount of detail. Looking forward to hearing what you think!
@nipun.nath & @siddhant.ranade from our research team will be on this thread tomorrow to answer some questions about Iris v2 and Nyx v1. I’ll make sure they get to this one
An issue with 3.5.0 is that in side by side view the source video does not show up until the live preview is generated. In previous versions the source shows up as soon as export starts. In 3.5.0 we have to wait for both.
@jacob.muchow
I can’t understand why you are rushing to release such a buggy version again.
Iris V2 introduce extremely “Motion Jitter”, which IrisV1 has not!
Unfortunately I can’t test this release any further, my Zotac RTX 4090 is now crashing under heavy load. RMA time. What a PITA! For a $2000+ card I expected better. $%#$@^@#!
I wouldn’t call Iris V2 “improved” but more “different”.
In my tests Iris V2 can give more details with medium to higher quality sources but is (sometimes vastly) inferior to V1 with LQ sources creating lots of funny things, altering peoples faces and introducing annoying repetitive patterns, oversharpening the background making the whole image looking unnatural/artificial.
So I think it would be best to keep both available, Iris V1 for LQ/MQ and V2 for MQ/HQ sources.
yes that’s true!
But the “Motion Jitter” is an extremely Step backward!
Jitter should be removed and not introduced (if it was even not there in the Source Video)
At the moment I’m interested in the W7800, actually not a good choice for A.I., but I multitask a lot, i.e. I often work on several things at the same time.
Here, my W6800 with two instances stutters when it has to do something in addition, but it works.
The W7800 should still have some room for improvement here.
Although the Quadro RTX 5000 (Turing 2080S GPU) and Radeon Pro W6800 are just as fast with one instance, the W6800 is 50% faster with two instances.
Or I’m thinking about getting an RTX Ada 4500, but it’s not available yet and the prices will probably take until December or March to become reasonable.
Test image with FFMPEG’s built-in testsrc filter. ver.3.5.0
Nyx has less color alteration.
For example, the red color changes from R:G:B=255:0:0 to R:G:B=254:0:0, which is within the margin of error.
Iris v2 has some strange color transformations.
For example, red goes from R:G:B=255:0:0 to R:G:B=238:14:11, which is fundamentally wrong.
If the color balance is made absurd, it will affect the grading process, so please consider an option that does not change the colors.
Other models can have less color alteration, so why can’t Iris do it as well?
)
).
