I just downloaded the latest build, installed with no problem and set my defaults. I added video to enhance, clicked process, and again with no problem. But what I didnāt realize, I selected a video I just spent 6 hours enhancing and the enhanced video file was overwritten without warning. Please add warning the target video is about to be overwritten, or automatically add a sequence number. I prefer the warning. Thank you.
I just noticed the option to use default output name, if unchecked, would have prevented the file from being overwritten, but even with that, a warning message should also be displayed before the process starts. Thank you again.
Could we have some infos about the GPU command line option ? what is it for and how do we use it ?
Meanwhile, the only two bugs I had with v1.2.0 are still remaining :
if I try multiple previews thereās sometimes a frame of the previous one displaying at the beginning
Remaining time when processing randomly not displaying.
About Artemis-MQ, there was real differences in v1.2.0 and it was often better than the LQ one who tends to go other the top with oversharpening and keep the compression artifacts from the original video. Itās like having a higher definition of the video that would have been encoding with a low bitrate.
GAIA-HQ is still the best but unfortunately canāt deal with very low definition videos <360p or videos with too much noise/macroblocks or poorly deinterlaced videos.
GAIA-CG is the best option for very bad quality videos but at the cost of a lack of fine details resulting.
If we could have a model with results in-between GAIA-HQ and GAIA-CG then this could provide amazing final outputs.
But for now Iām still waiting for the main mandatory new feature : Ability to save videos with any ffmpeg supporting encoding codec or vfw installed codecs on Windows.
Edit : Having saved v1.2.0 before installing v1.2.1 I can confirm after doing some tests than the old Artemis-MQ was producing better results on most videos than the new merged MQ-LQ model. Very disappointed on that choice to have delete the MQ model.
In 1.2.0, Artemis-MQ definitely did NOT look ANYTHING LIKE Artemis-LQ. Artemis-MQ was the BEST mode for DVD upscaling. It was SHARPER than both Artemis-HQ and Artemis-LQ in most cases, even on good quality video. Removing this mode is completely bizzare, and is demonstrably a DOWNGRADE.
Here are samples of video processed with 1.2.0. First one is Artemis-HQ, second one is Artemis-MQ, and third one is Artemis-LQ. Scaling is 2X from 720x576 source. After scaling, the Video Enhance AI results were resized from 1440x1152 to the correct aspect ratio of 1440x1080 using spline144 resize in AviSynth and a high quality crf 18 re-encode.
These are ffmpeg direct frame captures of the processed video.
I am not sure how you are testing your results to determine that there is ānearly no differenceā between Artemis-MQ and Artemis-LQ. I would recommend using the command ffmpeg -i result.mp4 -r 1/1 result-%06d.png. This generates one frame capture per second in the current directory. (If you want to extract all frames: ffmpeg -r 1 -i result.mp4 -r 1 result-%06d.png)
Here is a zippyshare link to all three files above for convenience:
Use this methodology:
Unzip the file.
Please install IrfanView. Use it to open 01_Sample_Artemis-HQ.png. Now press the space bar to open 02_Sample_Artemis-MQ.png, and again to open 03_Sample_Artemis-LQ.png. Press the backspace key to go the previous image, and the space key to go the next image. In this way, the differences become much more apparent and clear.
Here is the original sample clip used to produce the above results:
While it isnāt much, I am noticing things are running faster with the same settings in this version. I was averaging .65sec/frame, now Iām its .53. Doesntā seem like much, but doing the math, it appears to be shaving off about two hours of my encode time.
iām confirming this, iām running only on CPU i7-7700 unfortunatly, and my encoding per frame kind of been divided by 50%. i was at 12 to 15s per frame before and now iām closer to 4 or 5s per frame, no matter if the source is in full hd or not (before, the 100% deblocing was taking 15s per frame, while upscalling to x2 to x2.5 was taking around 4 to 6s per frame).
I have this problem where during processing using Video Enhance, when the computer goes idle (such as locking the PC, or going AFK), the processing usually stops. This wasnāt the case with Gigapixel AI.
Since this was still called Gigapixel AI for video, Iāve had problems with this software cutting out frames at the end of my videos. Since 1.2.0, the videos are the proper frame count, but the last frame of my videos is still missing, displaying a blank blank image instead. This is persisting with this release as well.
Really sucks when Iām upscaling CGI videos that are suppose to loop seamlessly. :\ The results are gorgeous, but it sucks I canāt use any of these videos.
Here are some video examples of VEAI not rendering the last frame in some videos. Would anyone mind running these through to see if they get the same issues as me? Theyāre 5 second, 304x208 CG videos.
I had a issue with that version related with the fact that is producing a ājumpingā video.
Recommendation is to downgrade it to 1.1.1 so itās look like that kind of bug in new version .
However Iām interesting is there any ETA for fixing this ? Iām now in trial mode but with such outcome I do not convinced to spend $ on it .