Video Enhance AI v1.8.1

If there is a file size limit then it should not allow me to upload. Period on that.

Ah I see what you mean,
We can consider to make those posts in the Facebook group as well.
That is a good point, we used to use that group as a BETA group, but since we changed the name of it, it does make sense to post regular releases there as well.
Thank you for your feedback, I appreciate.

YAY, thank you :slight_smile:

@cemal_gurel I need to make something clear. I love this product,respect a lot to the work that has been put from day 1 to this day.Yeah I sometimes got frustrated during the process but it has grown a lot during the period.The developers are working hard to bring the best of their effort,however that doesn’t mean it is a complete project.(There is always room for improvement and for new features)
Last year I asked for many requests and most of them are implemented as a feature(not all).I had a lot of ideas and still have more for better user experience but keeping the product working as intended was/is still a hard task in regards to the diversity of hardware,training speed is still slow to test the quality of models,optimization and for many more reasons. I totally understand those.That is why I respect a lot to the developers but that doesn’t take away the power of the userbase to ask for simple yet powerful options to be implemented.
I got really mad cos this is the 3rd time I lost all my models and I trusted 1.8.1 version is the latest one and 1.8.0 has already got the installer asking to keep the models.And it just deleted all :slight_smile:

My honest opinion:
1)We need small patches(not the full installer everytime)
2)If a feature is implemented do not take it back unless it causes issues
3)I always fought for give the power to the end users.Options over options.This doesn’t mean simplicity is less powerful but if you offer more people can achieve greater things with your product.Do not be afraid of results.
4)Do not delete our models please or limit the folder size etc to keep the hdd space low for the end users.If that is the destination deleting whatever etc,give users an option to choose to keep or delete in preferences.
Anyway accidents happen sometimes,it wasn’t a very pleasant one for me this time.Hoping for better experience on the later releases.

What is the best workflow to use for upscaling 1080p master > UHD. Can we use noise reduction (Neat Video) and sharpening on the 1080p file and upscale it? Or only noise reduction on the 1080p timeline and sharpen the upscale UHD version afterwards if necessary?

Is there any way to avoid color and exposure shifts? The colors and exposure shift after upscaling. Which is annoying if you apply upscaling on your master which is the final look you want.

Colorspace issues are only when doing SD>HD/4K conversion. HD>4K usually use the same colorspace BT709. BT 2020 is still rare.

I love VEAI, but I don’t trust it :grin:. I always back up my models folder just in case. I get how it tricked a lot of people in the last update, because version 1.8 just before it offered the option to keep your models folder. Sneaky!

2 Likes

We were using a couple of versions back before 1.8.x and the GPU processing worked on an i5 (2019) MacBookAir with 16gb RAM and Intel UHD Graphics 617 GPU with 1536 VRAM on Big Sur. This version crashes instantly on previewing or processing anything with GPU turned on. So now we are back to CPU only processing which isn’t reasonably usable on our work. Considering it was working before 1.8.x can you make it work again soon as we would like to be able to use the software we paid for
 particularly as we have seen its now discounted a further $60 or so from what we paid for it
 We need the software to reliably work on our machine and not start crashing in new versions as we rely on the software for our work.

Its also telling us when we start the app that our OS is out of date and not the latest version, despite Software Update telling us we are running the latest version of Big Sur.

1 Like

When I compare the lumetri scopes (before and after footage) I can see shifts in the colors and exposure. With my eyes I can also see the midtones, highlights and whites a bit ‘raised’. Tried to fix it with hybrid program but the instructions someone gave doesn’t fix the result


1 Like

Are you able to use models from 1.6 (and prior) in 1.8 releases? I would love to, because I’m not happy with the changes to Artemis models from 1.7 onwards. Artemis-HQ changed completely since 1.7 and not for the better, I think; it upscaled beautifully before from 480p to 1080p if you had a really pristine progressive source, producing natural sharp edges and none of the glitches existing in Gaia-HQ, but Artemis-HQ v8 (VEAI 1.7) had plain bad quality and v9 (VEAI 1.8) now lacks sharpness and fine details. It seems that the model completely changed, instead of making the 1.6 builds better.

Yes, I can install VEAI 1.6 separately, but it doesn’t enjoy the speed improvements made in 1.7, plus I’d like to have everything in a single place, if possible. Thanks.

1 Like

The only way to fix this is to use Avisynth and load the script directly into VEAI and output still images.
It’s just a basic avs script calling the video with the right source filter and convert the file directly to RGB with the right color matrix according to the source
Example

AVISource(“yourvideo”).ConvertToRGB(matrix=Rec601) for SD
or
AVISource(“yourvideo”).ConvertToRGB(matrix=Rec709) for HD

Then your PNG or TIFF will look exactly the same in terms of luma/chroma as the original source. Scopes will look different different because the video has more pixels and new pixels too but there normally will be no hue or gamma shift.

No, you can’t currently use models from 1.6 and prior versions. However, I was talking to the developer and he said he might adapt some of the older models to work in the future. As far as Artemis quality, that is something that seems to be very subjective. I am in the camp, along with many other people, that Artemis v9 is one of the best, most usable models Topaz has ever developed. To each his own, I guess.

2 Likes

Thank you

Is there any step by step guide on the internet that would help with this?

1 Like

Hi there, in the 1.8.0 BETA we did have an installer with a feature that lets you keep models that we are testing out.

We are currently still in BETA phase of implementing this feature. That is why it was not added to the public release.

Again, this feature is currently in development in the BETA channel, so you can use the keep models feature in the BETA installer for now.

If we determine that it is architecturally practical and of benefit to user after testing then the feature will be included in the public releases.

Thanks.

Hi there, we do not intend to deceive anyone.

An installer with the option to keep the models folder is currently in development in the BETA but this feature is not ready for public release into the stable development channel.

I hope that helps clarify a bit.
Thanks.

What universe do you live in? Every test I’ve done has shown ArtemisMQv9 beats anything prior, even GaiaCG 1.6.1 and Artemis MQ 1.2.0. The latest test is upscaling Aurora’s HAIK concert (clean 720p25fps) to 1080p50fps. The results are stunning. I actually backed off CAS sharpening to 0.9 rather than 1.0 as too much detail was making people look scary. The GaiaCG 1.6.1 version has the faint fine raindrop artifact problem, plus it can’t compete on details.

It will be 5 days before it’s done though. For the first time, the artifacts caused by SVPFlow are too severe to ignore. I’m having to use FrameRateConvertor which has no artifacts, but runs at 0.54 frames per second on my quad Xeon. My newest box with the 4.9GHz i7-9700K will run a short test at 0.81 fps, but I get a “corrupt video” error on the full length video- known bug with Virtualdub2 and Win10 (I need to see if VMWare or Virtualbox runs Win7 on 10).

Well the installer I have which was shared by Mr. Albert here,installs as release version and it has the feature to keep models.Also the product neither have a BETA shortcut,nor showing BETA in the product or else installing to the BETA folder path.Which means I have a release version that has that feature.

Hope this clarifies my frustration.

Thanks.

Edit: I am not allowed to paste a picture here.Check this link for the screenshot. I was so happy that day and I took a screenshot of it. 65.3 KB file on MEGA

Like johnnystar said, this is subjective and I prefer Artemis models from 1.6. Btw, I am only talking about Artemis, as Gaia-HQ v5 from 1.7 onwards far superceeds previous versions, I’m fine with that (except for Gaia-CG, which also produced sharper results before).

Artemis-MQ v9 is indeed good, I’d argue better than HQ, but it smoothens the image to a point that finer details are lost, and its sharpness is inconsistent. Unfortunately, Artemis-HQ and LQ v9 are much blurrier than they used to be, even though they do produce less artifacting (maybe because they use some sort of unsharpening filter? Beats me). I’d still pick Artemis-HQ from VEAI 1.6 and prior if possible, for upscaling from 480p to 1080p - but yeah, the source needs to be flawless for it to work well.

As for the “corrupt video” error with Virtualdub2, I suggest you use the x64 version if that’s not what you’re running.

1 Like

VEAI takes some seriously fast hardware to run on. What you have is relatively slow. I couldn’t find a TFLOPs rating on your GPU but found it has a TDP of 8 watts. You do know anyone serious about video upscaling is pushing TDPs of at least 200 watts? And the fastest are doing 300 to 450 watts (per GPU). I see this is for a business, umm
 spend the money and get a fast PC and GPU. I don’t even have a job and scraped together enough for two proper boxes for my hobby (I starve to get videos done).

I have a window air-conditioner cooling my two VEAI boxes. I turned the thermostat to the coolest setting so it never shuts off and have the compressor on low speed. I’m seeing 60C on the GPUs. The new box is running 71C on the OCed CPU at 4.9GHz.

Yeah, that’s the other workaround. I have to locate 64bit versions for all the plugins I use. I did a one day test of 64bit FrameRateConvertor on Avisynth+ and saw about 5% faster speeds. Hardly worth the effort. But that was before I got the 4.9GHz CPU. VM or 64 bit plugins. Hmmm