Hello everyone,
Another beta for Topaz Video AI is now available. Please test and provide your feedback. Thank you.
Downloads:
Changes from 3.0.4:
Notes:
Thatās great news!
Iām glad the Beta is still goingā¦ 
First time i see a working side by side comparison while exporting-good.
AV1 export using nvidia(rtx 4090) results in - Unrecognized FFmpeg option: ātilecolumnsā. Check the FFmpeg command.
AV1 export using intel crashes the app.Quality in auto mode looking good.Nice progress,although AV1 nvidia export was already working in 3.04-so proly a bug.
Great Improvement its quite similar to 2.6 version as upscale quality. will update after testing all models.
Will I finally get to use V3 if the quality mismatch has been fixed 
This is the first version of Video AI v3 Iāve used.
On a GTX 1060, a 720p to 1080p conversion using Proteus gave me an approx 30% speed improvement over v2.6.4 and the quality is about the same.
Ok - cannot preview at all.
On first run, Proteus default auto 2x upscale. Took about 2-3 minutes to start preview - the status ba will then start like its trying to preview, preview screen will still show generating and about 3x seconds later crashed the program.
On testing Proteus Auto to see how it fares without corrections, it still is a fair way away from usable on Auto only. Left is best result in 2.3, right is same sequence input in Proteus Auto in this Beta. I know its adjustable, but I think some more work on the Auto default to get it a bit better would be good.
Intel 12th Gen and above CPUs with integrated video also support AV1 acceleration. Will you also be supporting those?
Those CPU only support AV1 hardware decoding for video playback. At the moment, only Intel ARC GPU & Nvidia RTX40 have built in AV1 hardware encoder .
On my i9-11980HK I get about a 4X speed increase encoding AV1 using NEAV1 vs AMD or other processors. So are you sure about that? I realize what Intel is listing, but something is giving it considerable boost in encoding speeds.
The parallelisation seems to get better and better. Iām going from 0.88s/f to 0.55s/f if I split up the file and run 4 processes. Is there a plan to auto-split and combine a file to achieve that speed when processing just one file so that I donāt have to manually set the in/out trim points and stitch the segments together myself at the end?
Running multiple long previews in parallel leads to the system RAM being filled up to 100% and all processes crashing with āout of spaceā error.
Correction: any preview eventually fills up all available system RAM and crashes if length or resolution are big enough
I canāt login, āError: user not logged inā, although user and pass are correct and are working on the productive version
Interlaced Progressive means the source video was originally interlaced, but had been transcoded without proper de-interlacing. So, the interlace artifacts are baked into a progressive video because the 2 interlace fields are no longer separated.
Otherwise, the Interlaced Progressive model also works for most purely interlaced and purely progressive video sources, but (unlike the Interlaced models) does not double output framesā¦
Ok, so I know what kind of video file that is trying to work with.
My only concern is that this is a very specific file with issues that doesnāt exist without first having someone screw up the original file. This would mean the input file is a progressive video input that has significant leftover interlacing artifacts - but it is not interlaced, so the name is kind of weird.
Additionally, for most general users of this program, working out if this is the kind of file you have is extremely difficult for the laymen. If you have a genuinely interlaced file, it will look identical to this kind of file unless you look under the hood.
So is it expected the user is meant to determine the file kind first, or just experiment and see what works best?
Here is the comparison between MSI Afterburner and Taskmanager for the workload detection.
The task manager does not even show half of the half although the GPU is fully loaded.
As I said, it also works for both true interlaced and progressive videos, so itās quite versatile. Personally, I find it retains more detail than other models, but it sometimes creates artifacts of its own.
In versions 2.x, the output window is in step with the input during processing. (and during preview)
In 3.x, output pauses, runs, pauses. I for one do not like that at all.
Could you change that? Iāve commented on this before, several times, but Iāve never had a specific reply!
Perhaps itās impossible to do now, since the software has been rewritten completely, but it would be nice to know if youāre looking to change it.
Yes, not being able to compare is a deal breaker for many.
