Is it the same with mp4 → TIFF 16 bit uncompressed?
Sorry but how do I get Hybrid?
Load RGB inputs in VEAI and the colorspace issues will be solved. So how do you do that :
create an avisynth script and add at the end of it :
For SD videos
ConvertToRGB(matrix="Rec601")
For HD videos
ConvertToRGB(matrix="Rec709")
What video card do you have?
Spacedad,
What video card are you using with 1.7.1?
2080ti
Is VEAI supposed to keep subtitle tracks that are present in the input file? When I process h.264 mp4 files with VEAI with mp4 as the output format, the output file contains no subtitle tracks, even though they were present on the input file.
1 Like
No. Mkv or other Contener having subtitles tracks are encoded in Mp4 or Pictures sequences so the subtitiles tracks are lost.
Thanks. Didn’t realize that.
Just use MKVextract to extract your subtitles file then remux with MKVmerge.
No, 10XX can’t process fp16 like 20XX and 30XX do.
VEAi has been a great succes and does a great job and is getting better with each update.
Sadly there has been no effort made to make the process of creating sequential still images any more convenient since VEAi has been much improved over time. Here is the issue with the current way things work.
1 Once you mark the end and start frame and do the processing, you have to un-check mark the video file and then when you want to mark the next sequence you are forced to first mark the end and then the start file. This is more than annoying.
2 There should be a way to mark more a sequence of non sequential group of frames and then have them process all at once. This would be very convenient and save tons of time.
A Slightly Different Matter:
The most common expert advice for getting the highest quality video upscaling suggests creating still frames of the entire video, processing them as stills (with DeNoiseAi, SharpenAi, and GigapixlAi) and then converting the processed/upscaled frames back into a single video file.
However, there is a problem here. My experience is that the still frame outputs do not process very well either with DeNoiseAi, nor with SharpenAi, just like Scans of printed images. Perhaps this is because the AI was never focused on working with such images.
For the purpose of making the conversion of groups of non sequential stills back into video format,
it would also be very handy to have these sequences numbered sequentially - because when converting the stills back to video, frame numbers should not be broken up.
Thanks
1 Like
I am getting odd horizontal artifacts from Artemis V8 HQ. I can always tell they are an artifact because they have this top facing notch as well:
It’s like the model is processing the image in sections and not properly piecing it together? They are very random, and the same frame in a different file won’t show the artifact, but then they will appear elsewhere.
TIFF is so large you’ll need A LOT of storage for even a short clip. Whether it happens or not, not too sure. It definitely happens when outputting to MP4 or PNG.
Hey guys, I just upgraded from a GTX 1070 to a RTX 3070 and I’m a bit confused.
I made a series of tests with the same video (640x400).
Upscaling it to 1080p the 1070 needed 8min (0.47sec/frame) and the 3070 3min (0.22sec/frame)
Upscaling it to 4k the 1070 needed 8min (0.51sec/frame) and the 3070 6min (0.45sec/frame)
The 1080p render showed me what you would expect, 2x the performance of the 1070. But the 4k performance and GPU usage I find strange. Using HWMonitor it showed me for the 1080p render a usage of nearly constant 80%, but for the 4k it never got over 70% and often fluctuated between 30-50%.
Do you think this is a problem with the current drivers or a software issue?
just wanted to ask again… is there any updated documentation for the command line options?
would love to be able to specify ProRes for output (HQ seems overkill too, so can we access other prores profiles through the ffmpeg integration?)
how do you properly specify, or can you even, multiple GPUs in the command line similar to the “All GPUs” option in the GUI?
Thank you!!
1 Like
How do you enable OpenVINO in VEAI?
Can any developers confirm if VEAI will see a return to the quality of the 1.6.1 models? It seems developers are sending a message that they are aware of the quality drop, but they can’t implement 1.6.1 models due to “technical reasons.” So there is speculation that 1.7.0 and forward is trading off quality for a speed boost.
Users can only know what they see, so I think it’s important for some light to be shed on this situation, rather then letting users’ speculate and letting rumors spread.
1 Like
It’s never overkill until it’s still lossy. I want 4444 XQ. I’ve been testing ProRes vs PNG and and there is a visible difference even at 100%. It’s overkill only if you plan to reencode the file as it is for consumer level but if you plan some postprocessing better to work with the highest quality possible.
Have you tried in the CLI
–output-format mov
by “visible difference”, which one is better quality?
