If your video has more than 100,000 frames change %06d.png to %07d.png
Any plans on adding a preview mode which previews all ai models to compare at once?
5 Likes
Great idea! We want that!
Also sometimes, mostly with HD videos if you plan a 100% denoise/deblock you often will have better results by first downsize by 2 your file (spline36 resize with a lossless temporary file, for that part I donât even anymore bother with temporary files because I use Avisynth virtual file system) and then do a GAIA-HQ 200% instead. And the bonus is that itâs even more faster to process.
1 Like
Already done that to a seriously noisy shot on a wedding video that even NeatVideo couldnât fix 
We should get all those âhintsâ on a list, should be usefull to some people, we all lost hours and hours of experimenting with videos so far.
The program has a bad habit of automatically resizing video files to an odd number of pixels vertically, horizontally, or both. This is simply not acceptable for video files and can result in all sorts of strange behavior and crashes in video players and editing software. Iâm guessing thereâs probably a lot of bugs being reported that can be traced back to this single issue.
This tends to happen a lot with video files that have non-square pixels, because the software first has to convert them to square pixels for the AI to use. The software is being so strict about maintaining the exact aspect ratio of the image that itâs forcing these illegitimate resolutions. This really shouldnât be happening, or at least it shouldnât be the default behavior.
Even if the output is set to be an image sequence, which can technically support odd resolutions, the end result is always going to be a video file. Users are going to have to resize them to legitimate resolutions anywayâand thatâs only assuming they understand that thereâs a problem in the first place.
Youâre right, i have this issue with Mp4 H264 file subtimed in 768x576 (square pixels, they are converted from interlaced Mjpeg Avi â de-interlaced â Mp4 H264 progressive) â VEAI , fortunatly, the issue is not a too much big deal because iâm doing a widescreen conversion of a restaured movie, but when the pixels is lost on the right or left part of the picture, itâs unfortunate, and I must crop each files.
same with the lost of 1 or 2 frames which happen sometimes on almost all videos converted i must put these frames in a separated video, and put some empty frames at the end to have them processed and fill the missing parts.
Hope to have this fixed at some points. but i still love very much this software despite al this ;-).
Keeping the exact right aspect ratio is a good thing. If you are processing anamorphic videos then never output to mp4 but in still images. After that you can on your post process add black borders to match the mod your final codec requires.
@steven.guy ⊠How is PAL affected by pixel dimensions? I ask because I purchased full series Star Trek Voyager in PAL format, thinking the slightly higher resolution would result in higher quality upscale, but it didnât. In fact, it was worse when I compared the same episode to the NTSC version, which used the same rendering model Gaia CG.HQ 4K. Now Iâm in the market for cheap Voyager NTSC set.
I donât know the specifics of that conversion, but Iâm going to have to assume due to its age that the show was simply converted straight over from NTSC to PAL. This means that even though the PAL version has âmoreâ pixels, they were just artificially added as part of the upconversion process, from whatever old machines/software they used back then. If you do side by side comparisons, I would guess the PAL version is blurrier and may have weird fringing going on in little details.
And even if they did actually go through the trouble of rescanning the source film and re-edit everything just for PAL (unlikely), theyâd still have to do something to change the frame rate from 24 to match PALâs 25fps, and the processes theyâd have available do this usually werenât all that sophisticated. In fact, that process is arguably even worse than the upscaling.
So yeah, always stick to whatever the original broadcast was. Otherwise, youâre going to be giving the program footage that has a bunch of artifacts in itâeven if it looks okay at first glance.
To maintain 1:1 square pixel before converting in Video Enhance AI, you can convert to mkv using MKVToolNix. This doesnât re-encode the video.
Set the âDisplay width/heightâ to the exact dimensions of the video, for example 720x480:
Otherwise, you can re-encode before going through Video Enhance AI using FFmpeg or Hybrid. Set pixels to square, set CRF to 1 for close to lossless.
Yep. For me on 1.2.1 it was dropping the third-to-last frame, then came back for a single frame, then dropped the last frameâŠthose two would just be black. And it seems like itâs fully removing several frames from the end beyond that. Now with 1.2.2, itâs also removing several frames from the BEGINNING of my video as well. On a 2017 MacBook Pro, OSX Catalina. Iâm processing very short clipsâunder 30 secondsâso perhaps the amount itâs removing is proportional to the length? Weak sauce.
1 Like
Absolutly the same issue, iâm actually restauring a short movie i worked on 15 years ago, we did (with the help of someone on this forum, thanks to him, as my graphic card is too slow) the upscaling clip by clip and not on the whole movie (35mn long) because i had several different sources for the restauration.
unfortunatly itâs missing 2 to 3 frames to a lot of clip.
fortunatly some are unafected (why ? no idea). it seems i didntâ noticed the missing frames at the beginning because i re-edit a lot, iâll hate to check that ! hope they will fix that soon ! unfortunatly itâs a 1.2.x software, so itâs unfortunatly normal today that such bugs happen even on 200$ software (thatâs worst on 700$ software lol).
they are released kind of in a beta state, to earn money quickly to continue to produce etc⊠thatâs the norm today⊠we are beta tester without asking it lol.
the good point : the software is released now and not in 2 years but there is a counter part to it.
Hi,
Are you planning to add NVIDIA frame rate smoothing Super slomo its using same cuDNN system you can use this to increase fps for upscaling videos or frame smoothing for more smooth transitions. Also i was wondering if you plan on adding optimization for empty/simple frames like full black frames in transitions, currently upscaling black frame from 720p to 4k takes same amount of time as frame with way more information. 
Amazing work on current version!
I read somewhere else on this forum that the lower the resolution the better the upscale & it so so correct.
I was upscaling the colorized version of Casablanca (1942) to 1920x1080 & thought I would update it to 1280x720 with another video editing software as this would give Topaz an easy job but was shocked when it said it would take 30 days (yes, 30 days to do) I think it was over 5 seconds per frame.
So I put on my 640x480 version & it dropped the frame rate to just under 1 second & the result was so good.(or as you said âstunningâ
So now I will be using the 640x480 method on all my old movies.
Will there be support for AMD devices? CPUs and GPUs?
4 Likes
Itâs a bit of a guess at this point, but I suspect the missing frame(s) at the beginning of the output video normally occurs when processing a clip from a longer video, where the first frame to be processed occurs in the middle of the GOP structure.
It might be that correct processing can only begin once the video has reached the next I-frame (keyframe) after the specified start point. As far as I know ffmpeg can do frame accurate seeking (the timeline slider works ok) so this shouldnât be a problem and should be fixable (assuming thatâs the issue).
I also would very much like solutions to point no.s
#3 I 've had power outage 3x times while processing a video that would take several days. It would be excellent to have some automatic storage every few hours, so that the program could continue from a crash.
#4 Seek function sometimes moves too slow or too fast. Using the L and R cursor usually moves frames, but it often refuses to function until you use the frame by frame arrows.
How do you downsize by 2 using spline36 resize? Is that possible with command line only ?
If yes could you give an example ?
Thanks
I do it with an avisynth script, then I donât even have to encode the new resized video. I just mount a virtual uncompressed AVI file with Avisynth virtual system file aka AVFS. There is a 64 bit version included with vapoursynth but the 32 bit normal version should be OK and prevent out of memory crashes for a simple resize script.
