Yes it looks better Lanczos is not something I’m very familiar with but my goals are
1 - upscale to 4K to watch old TV shows on bigger screens (65, 75, 85 and even bigger someday) In a way for the children and maybe their children in the future, who know (perhaps even 8K will be outdated with the holographic TV
2 - Improve the quality giving that it’s not altering the natural aspect of the human beings. I don’t like to plastic look of some converted videos I’ve seen here and there.
3 - Smoother background , more natural details on any objects or face, less blurry.
Artemis LQ does just that so far. But only one episode has been upscaled / enhanced. I’ll post the snapshots I"ve been taken soon.
Have any buffer in this AI?
Because I tested with a few 3GP files, and 600% zoom. Second encode was OK with Artemis HQ, but when try bigger filtering (more noise and sharpen) after two minutes program crashed and my GPU goes black, I need turn off my PC. If try bigger video, then crash comes after few seconds. This great AI, but I’m sad why can’t use with my GPU (AMD RX 460).
Try setting MAX VRAM to the lowest setting and select Reduce GPU Load. If you still have issues can you message the VEAI logs?
To activate logging Help->Enable logging then restart the app.
Do the normal processing to crash it and after that.
For the logs go to Help → Open Log Folder and zip the entire folder.
There are three bugs I’ve discovered with the latest Video Enhancer AI v1.7.1:
(1) not all the GPUs are shown in the Preferences.
I have two RTX 2080 TIs in my system, in V1.6.1 it was fine, both GPUs were shown and selectable to run as individual instances of VEAI. Now, only GPU1 is shown, GPU0 disappeared.
(2) When GPU1 was selected as the one to use, it turns out that VEAI is still using GPU0, even after restarting the VEAI.
(3) When All GPUs is chosen in Preferences, VEAI still only use GPU0, and GPU1 is left un-used
I’ve been stuck using v1.3.8 as the newer updates where all about half the speed or made workflow slower.
Finally updated to v1.7.1 after hearing about the improvements. Updated to new Studio Drivers and the 20H2 windows updates.
… and
I was on my Ryzen 2060KO machine.
Artimis LQ-V8 doesn’t even bring the GPU off idle on any setting and only uses 700mb of VRAM and 21watts.
And encodes 1440x1080 video at ~1.68 spf
Tried all the settings, none changed anything.
Gia v5, encodes at 0.18spf and used 100watts.
Something up there.
Topaz 1.3.8 does it at ~0.49spf
Same results on my laptop with 2060. Both Artimis V8 are broken slow. I’ve been really liking Artimis for bad old DVD style content. But this is just broken. I suppose, I’ll have to give up.
I’ve just enhanced a 720p movie to 1080p with Video Enhancer 1.7.1 and also found a bug. The upscaled movie no longer plays smoothly - there are micro stutters.
I’ve analysed the output frame by frame, and frames are sometimes doubled and the next one skipped.
Frames original movie: 1 2 3 4 5 6 7 8 …
Frames upscaled movie: 1 2 3 4 4 6 7 8 (4 is doubled, 5 is skipped)
Windows 10 up to date, latest nvida Studio driver and RTX 3090 FE. The 3 hour movie was processed in roughly 13 hours - 0.19sec per frame - and besides the stutter, the output looks very good. But it seems there is a bug regarding correct frame processing (at least, with these settings).
Usually a good model can be found to match a video, but sometimes the effect is too extreme. It would be nice to have an adjustment level, something like 0%-100% model. Sort of like wet/dry in an audio filter.
That’s incredible speed with the RTX 3090 FE! I’m hoping to eventually get a RTX 3080 or AMD RX6800 XT. It seems every store has no stock for them.
Is your movie source is .mkv or .mp4? It’s progressive video right? What’s the framerate? You may have to use DGDecNV (very well written H264 source plugin for Avisynth, $15) to get a stable video source.
Is Gaia-CG working or not (like it did in v1.6.1)? In the release notes for 1.7.1, the developers note that Gaia-CG is not working properly. Somewhere in this thread, someone mention it is working correctly. Does anyone know? I’m running v1.6.1 + GTX1080Ti and getting excellent results with Gaia-CG, but would like to get a new faster GPU supported by v1.7.1. I’m avoiding upgrading due to the Gaia-CG issue.
Source is .mkv. Frame rate is 23.976. And progressive of course Currently I’m re-encoding the 720p source in the hope it will be a more stable input for Video Enhancer.
In the meantime I’ve tried Video Enhancer with another .mkv source and that didn’t cause frame issues. So, it seems source related or combination of source and Video Enhancer.
Yes, with Gaia-HQ the 3090 FE is properly used: 70-75 constant load. When I do an upscale of 1080p to 4K it is even used more: 80+% (0.43s per frame)
A re-encode of the source seems to have fixed the issue; tested a scene with Gaia-HQ and no funny frame skipping. Good, will start again a 13 hours process.
Update: this time it worked perfectly VEIA probably had issues identifying all discrete frames of the source file. After a pass through Handbrake, the source file posed no longer any challenges for VEIA. Lesson learned: upscale a sample and have a very good look at it first (I did a test sample but was too focused on upscale quality), before starting up a very lengthy process.
@virtutis the message you wrote about upscaling Star Trek Voyager seems to be missing and I’d like to ask you if you can resend me the QTGMC settings you use for it
My experience with Artemis and Gaia parallel yours to a point, but not nearly completely. Artemis barely ramps up the GPU. Gaia ramps it up a lot!
But! For me, Artemis is easily twice as fast as it ever was before, and is much faster than Gaia. I don’t know what is making yours slower. 2080ti here. I love the faster with lower power that I’m seeing.
Well, when the model is 4 times slower than previous software release. I assume there is a problem. Any normal person would call that broken on the same hardware.
The Nvidia Studio Drivers and Topaz 1.7.1 notes stat that Tensor Cores will be used. I have not been seen any evidence. Which steps are needed to get Tensor cores used?
As I pointed out, mine is not 4 times slower, but rather 2 times faster. So the question is what the difference is.
Did your system download the fast models? As noted, currently the fast models for Artemis aren’t available, so if you never go them that could be a factor.
Lanczos is not something I’m very familiar with but my goals are
Currently I’m re-encoding the 720p source in the hope it will be a more stable input for Video Enhancer.