Video Enhance AI v1.9.0

I have the same exact problem with my PAL 25 fps videos. If I use a deinterlacing with frame rate doubling the audio gets out of sync, if I deinterlace with no framerate doubling the audio is fine.

Hey everyone! A bit new here and I have a question.

I run a pretty successful YouTube channel and I’m going through and reviewing some older sci-fi shows. I’m wanting to know the best way to really clean up footage for my reviews to REALLY pop. The show I’ll be using is Sliders. I have the DVDs burned and have been using the files. Some of the clips look good. Some aren’t. I’m really wanting to make a difference and add an additional quality to my video.

Could someone give me the easiest/quickest route (if possible)

Put the clip into Topaz 1.9.0
Scale 225% to make it 1080p
Use the DTD1 Model assuming it’s interlaced.

Just an unhappy report.
Topaz 1.9.0 crashed overnight on all three computers. The ungraceful crash where it just disappears from the task bar with no message.
Haven’t had such happen with Topaz since Spring.

I am in the trial period and so far very very impressed with the results with the new Dione models! My use case is very specific - enhancing recorded output from mini and micro analog cameras (much of it recorded wirelessly) in NTSC and PAL formats.

How can I make an informed decision about ā€˜field order’?

Also, with noisy low light recordings, should I do a noise removal step first? Same question for electronic interference noise [typically exhibited as diagonal or horizontal lines.

Exciting stuff!

I’ve run the DVD version of Sliders through DTD v1 at 1080p.

It’s coming out a bit weird to me. A lot of artifacting all over the place. I have yet to find one I really like. I’ve seen a lot of YouTube videos online of upconverts and they look pretty damn good. Almost everyone I’ve done either has way too much artifacting, creates a painting like effect or just doesn’t work.

Hi,

i just tested the 1.9.0 with interlaced dvds (adventures of young indiana jones)
First i used DioneTV V1 and the Picture was De-Interlaced very well. What is really amazing.
Ok on other interlaced sources that sharp ā€œsaw toothā€ were not perfectly deinterlaced, it was more replaced by soft ā€œsaw toothā€ like a sine wave. But also here it was way better than the saw tooth original.

During playback i pretty soon realized that all episodes of ā€œyoung indyā€ have been completely out of audio sync. Some even with a 20+ seconds delay (either too early or to late)…
I know that audio was not fully supported by VEAI, at least i did not read since …that it is now.

I already upscaled lots of different sources with different models and i never had any issues with audio.

Now, after i tried the DioneTD v1 on the same dvd source (indiana jones)
i can tell this modell has NO ā€œaudio out of syncā€ issues. All is lipsync and is played back perfectly.
So i recommend to use DioneTD on SD/DVD videos until this audio issue on DioneTV is solved.

You had success with DTD v1? I was trying to do it on Sliders and I wasn’t having any success. It looked horrible.

Have you used mediainfo to see that Sliders is a ā€œtrueā€ interlace, and if so if it is top first or bottom first? Lots of ā€œinterlacedā€ out there are ones that have been badly converted to progressive, and if it is true interlaced you have to get the field order correct.

THOUSAND OF MESSAGES!!! AND!!! NOBODY FROM ADMINS DIDN"T PIN USERS POST WITH INSTRUCTIONS. IM SCREAAAMING!!!
ADMINS!!!
PLS MAKE A THREAD WHERE IS ONLY TUTORIALS/METHODS FROM USERS. AS A NORMAL USER I LETERALLY DON’T KNOW WHICH METHOD ARE THE BEST.

This might sound harsh, but

There are no best methods. Even the professionals around here have to go by trial and error.

With every video it’s the same process: Try each model. If the source is interlaced, then you go by all the models with deinterlacing, if you are not happy with the deinterlace models, you go by QTGMC (and there are no best methods for that one either. It’s try try try) and then you go all the models again…

Sometimes it’s best to go 100% denoise and after that upscale, or use Neat video for noise removal, and then upscale.

Learning is never ending.

1 Like

First make sure to have a proper extracted source. Is it NTSC or PAL DVD ?
If it’s NTSC make sure it is detelecined and have a 23.976 fps progressive film source.

I would try Artemis LQ V8 that was really good with DVD sources. To make it available on v1.9 go to the
C:\Users\yourusername\AppData\Roaming\Topaz Labs LLC\Video Enhance AI\models folder
open ā€œartemis-lq-10.jsonā€ file with notepad, change in the 3rd line the version number from 10 to 8 and then save the file as ā€œartemis-lq-8.jsonā€. (make a backup first of artemis-lq-10.json in case you overwrite it by mistake)

When you next open VEAI, you should have Artemis LQ V8 on your models list, if you choose it and just launch a preview it will automatically download the right V8 model ready to use next.

Over the Facebook group someone created an installer that features older models.

Can’t find anything using the search function, so I guess I’ll toss the question here even though it’s probably the wrong place.

Got my RTX 3080 crunching away and even with that powerhouse I’m looking at about 6 seconds to convert 1 second of video.

My question is has anyone considered programming an FPGA for this? I remember back in the early days of Bitcoin when I went from my GPU rig to my first purpose built FPGA and I got about 10x the perf at 1/10th the power. My numbers might not be exactly right since I’m going from memory, but the gains of having purpose built hardware were dramatic.

I can only Imagine if Topaz got some engineers set on this task they could diversify from just providing software to also selling FPGA’s with dramatic improvements in throughput and probably make a pretty penny on the hardware sales besides.

The ones that are the best are the ones you find yourself and look good to your eye.

1 Like

Video Enhance AI 1.9.0 is changing my resolution from 740x480 to 640x480 when my scaling is set to 100%.

Anyone else have this problem and what did you do to fix it?

I checked. It’s progressive. This might explain why the Diode doesn’t come out looking great. So far, the best looking is AMQ in my opinion. Best detail. Decent sharpness and doesn’t ā€˜stamp’ the image down.

Your file certainly have a display aspect ratio set to 4:3 , you should see a value different from 1 in the SAR display on your VEAI interface.

Unfortunately, the Dione models aren’t able to do anything about interlaced videos that have been badly converted to progressive. I’ve had varied success using QTGMC with a filetype of 2 or 3, but varied is anywhere from passable to no improvement at all.

Interesting. I’m certainly seeing an improvement, but I don’t believe it’s as Earth shattering as some results I’m finding on YouTube. Which sucks. I was really wanting to have this added element to increase the quality of production. Not a deal breaker.