Video Enhance AI v1.8.1

Agreed, that’s not a functional change…it’s just data.

Thx for the info

i am waiting for a new/hopefully better Gaia Model, since IMHO this model ist best for very old low resolution videos like 640x480 or even lower.
Gaia sharpens them very well, but then there is much skin disturbance left.
It looks too hard

All the Artemis models are IMHO pretty useless for very low resolutions, since the results are over softened. Skin looks too soft/wax like

Id like to see a Gaia like Theia model
where i have controls to change sharpness and deblocking and softness

1 Like

did you try ‘2-passing’ the video? First with Artemis then Gaia?
There will be never a ‘good for all’ model that you can just throw any video and it will magically enhance sharpening, upres, deal with artifacts, double framerate and workout color space (and I guess there may be another hundreds of things people will want to do with the software). That’s the nature of AI. You train your algorithm and the better it gets in doing something, it also gets more and more specific for that type of source you’re feeding into the training program and/or that specific task you’re training. A huge neural network that has every branch of various algorithms and ‘decides’ which one works best for your sorce would be the perfect scenario but I guess we’re not anywere near that (at least for ‘video enhancing’).
That’s also the reason why we need to be able to maintain older versions of models (at least the ones that work with the current engine).

1 Like

Is there any chance we could skip the “Do you really want to quit?” dialog box? It’s not necessary. We click on the close button for a very specific reason that we shouldn’t have to confirm every time.

In my two systems:

10900K
720P to 2K(200%) Artemis v9
RTX2080:0.22~0.25sec/frame
RTX3070:0.22~0.24sec/frame

10700F
720P to 2K(200%) Artemis v9
RTX3070:0.34~0.36sec/frame
RTX3080:0.33~0.36sec/frame

So what’s the point of upgrading GPU?

3 Likes

I would like the option to keep all frames. VEAI always skips the last 1-3 frames of a video which is fine for most videos but right now I want to upscale a bunch very short gif like loops. Unfortunatly skipping even 1 frame breaks the transition thus ruinning it.

These are old perfect loops and I don’t have the original sorce material anymore so I can’t just properly keyframe them so i need to work with what i have. thanks.

Photoshop also asks, but only if any actions were performed. Otherwise, there will be no additional questions.

Hmmm, you could try something on the 30XX Series.

Start the Nvidia Settings APP.
Go to → Manage 3D → Programm Settings → Add → VEAI.
OK,

After you added VEAI
—> Power Management = Maximum Performance.
—> Latency = Ultra
—> Threaded Optimisation = On

After you did press apply and closed the window you press five times F5 in your Desktop.
And then you start VEAI.


But i think this will change nothing. :crazy_face:

You have very good data for analysis.

So I forgot:

Please list type of RAM (full specs), Mainboard & OS and number of GPU driver.

Update:

Looks like the Processor does break.

Any AMD systems here with 30XX GPUs?

Try again but with GaiaHQ instead since it’s more gpu bound than all the other models.

Sure, if there is a major encoding job currently underway then it makes sense to confirm. But much of the time I’m just briefly launching VEAI to test a random video and then exit. I don’t want to be constantly asked if I know what I’m doing.

win10 20H2
nvidia studio driver 460.89

Gigabyte Z490 AORUS MASTER

DDR4-3200 Corsair 32G

On both systems?

I need type of ssd also.

During heavy workflow and a dozen of open screens you easily can hit the closing X

It happend at least 5 times to me
and without that dialogue i would have lost an hour of work on a new upscaling queue

So better the devs let it where its is

1 Like

well i thought about
but as it also doubles the already long enough rendering times, i did not test it yet

how do you upscale "difficult sources?
first with artemis directly to 4k? and then gaia again on that artemis 4k vid?
or
do you first use artemis up to 1080p
and then use gaia for the final 4k upscale from the artemis 1080p vid?

i even found out,
when i have a bad 720p/1080p source vid and upscaling this source is not delivering my prefered results, i pre-process the vid in adobe premiere, without any filters/enhancements,
just saving it in 480p or 360p

and then i use gaia to upscale that now much smaller file gives me often much better results…
IMHO

but … i know, thats always a question of taste said the monkey and bites the soap…
what i think is amazing, may not be amazing for others

I like that hard to explain 70/30 look after using gaia on very small vids like 320x240
what means the upscaled vid is 70% unsharp and 30% sharp during upscaling
but not anymore during playback

Our brain (at least mine… lol) is capable of “correcting” the unsharp parts, as they change very fast,
sharp unsharp sharp unsharp … so the whole vid during playback looks kinda 100% sharp with a small VR 3D touch… its really hard to explain and sounds like nonsense … but it works for me and i really like it

c’mon
I purchased the RTX3090 Founders edition in a very lucky moment for the original price of
1499$ in december.

And yes, VEAI does not use it to the limits → on one VEAI instance.
But on 2 or 3 parallel for sure… and then the performance is really good…
I did not regret my investment
Maybe i would take the cheaper 3080 today, as even 3 instances of VEAI hardly use 10GB Ram.
But as there are only totally overpriced 3080ths available … thats out of question

Question about the ProRes implementation: Is it an officially licensed encoder from Apple or is it the FFMPEG version? It’s not a huge deal if it is, just might need to run the completed video through Compressor an extra time before turning in a video to a studio or broadcast network.

I had projects for which the second pass took more than 2 weeks

The maximum I go is 200%. If source is 720p, I use 200% (to 2560x1440) then the next pass to 5k. Then if needed I use Shotcut or Premiere to adjust color/tune and render it to 4k.
In which order you’ll use the algorithm (if first Gaia, or Theia, or Artemis and each of their ‘flavours’) is up to you. To decide that I sugest you only use a small portion (less than 10 sec) of the video to test. This way you don’t loose too much time on testing. I usually go first Artemis-HQ then Gaia-HQ, but I always test first.

Sometimes it is better, if the source is very damaged. Sometimes by doing a 100% first pass you can improve the image and get better subsequent passes. Again, you have to test. There isn’t a ‘magical receipt’ that will work on every video.

1 Like

thx for the infos, i will try that 2 step rendering asap

hi I’ve been using the trial version of 1.81 and im loving the software and I’m considering to buy it but recently I’ve been getting this “error” that’s been ruining my experience. if I try to scale a certain mp4s over 200% (like in 1920x1080) in any ai model the software gives up and just spits out “information: unable to run the selected model” I know my system can upscale to 1080p as I was able to do it with no problem with another mp4 file. I even made a new version of videos that got this error in my editing software and VLC thinking it was how the video was encoded to be the issue but after multiple tries I kept on getting the same message. is there anyway to fix this certain issue? I’ve even reinstalling the program with no luck.