- Command Line Interface Revamped the command line interface to make it easier to use and created full documentation
- Theia Model Update The Theia model has been updated with a new version.
I believe there is a bug in the new 1.4.2 VEAI. When you select the GPU and the program prompt you to click restart the program will actually shut down instead of restart and you have to launch it again.
What the hell is this version? There is no “keep audio” settings and default output JPG . what?
Hi there! The keep audio option is shown under the section box to select mp4 in the side panel for easy access. The default was changed due to long standing issues with the mp4 output (incorrect color, audio sync, bad quality, crashes, etc). While we work on those we highly suggest using one of the image sequence outputs.
8KUHD preset still isn’t fixed. The mp4 output is still corrupted after the process has finished. Please fix this.
I believe the codec actually doesn’t support 8k. Try using image sequences (jpg should be good enough if you are used to the quality of mp4). Perhaps we should put a warning that it won’t work.
Image sequence works fine. But why does the 4KUHD preset work with mp4 then? 8K is just a step up from that.
You would have to ask the implements of the H.263/4 standards! I imagine they thought at the time no one would need video files larger than 4k. H.265 does support this but out video IO pipeline does not. Unfortunately until that is redone you will need to stick to image sequences for 8k. Plus the program is far more stable using those!
Good to know. I’ll stick to image sequence for 8K then. Thanks!
What was actually changed with the Theia model? Wish you guys would give more of an explaination.
I couldn’t agree more! The reality is no parameters changed or anything like that. The model simply had more time to train and “improve” based on the metrics we set for it. It got (in objective terms) closer to a perfect upscale of video (as in looks closer to the truth). So when we say “Theia model improvement” it is likely that we changed very little besides steering the training away from certain problems (or in mathematical terms local minima) in search of the true global minimum of upscaling videos. We will likely write “X model improvement” on every one of our releases because we always have our AI training machines running trying to improve and we just take the latest and greatest each time we release to include.
I’m quite certain this is a far more technical explanation than you were looking for! That said I hope it helped a little bit in understanding how AI works.
Thanks for the reply. I was wondering could the future release notes include more info about what kind of expected improvement is expected when it say “new AI model” or “improved AI model” etc. Having some kind of sensible explanation about what improvements are expected would be of great help so people can know is it worth installing new version and redoing old projects and do tests. If you could include things like that in release notes that would be great, thanks.
Also sometimes I noticed new releases but release notes only show up later. Is there are reason why they are not released at the same time?
Il programma, come le precedenti versioni, continua ad essere terribilmente instabile con qualsiasi modello e va quasi sempre (95 volte su 100) in crash quando si effettua un salvataggio in mp4 di un file (qualità DVD portato a 1920x1080) con molti fotogrammi (3000) ed usando la GPU Nvidia RTX2080TI. Tutto OK utilizzando la CPU Intel I9. Ovviamente i tempi di elaborazione sono molto diversi fra utilizzo di GPU o di CPU con un rapporto di 1/10
Hi Ashley Tharp, Matt Lathrop, Beta/release version 1.4.2 has issues with “-f 16tif”, “-c x” and -o [image extension added]. I love the "-o " output directory option; now all four GPUs can output a video segment to the same folder/URL albeit with a bug covered below.
C:\Program Files\Topaz Labs LLC\Topaz Video Enhance AI\topazvideoenhanceai_cmd.exe -i “\192.168.20.155\media\Movies\test.avi” -o “\192.168.20.155\media\Movies\test.avi_Scale_x2_Gaia-HQ” -f 16tif -s 2 -m 2 -c 2 -b 1000 -e 10000
Issue with command-line options
-f 16tif outputs to file type jpg instead.
-c 2 Changing GPU doesn’t work all -c x settings 0,1,2,4 use GPU 2 on my system.
-o “\server\share\folder” Is mostly ok except VEAI doesnt use specified folder (existing) VEAI creates a new folder with " " --> “\server\share\folder_jpg”
These options work fine
Don’t know if it is reported or not. There is a bug in the program that near the last 1 or 2 seconds of the processed output, the video will repeat/loop itself. I tested several short clips (10 - 15 seconds) and all have this problem. Other than this, the program is good. Just use Theia Fidelity model and set Restore Detail and Sharpen parameter to 100 will give very good 4k upscaled result for 1080p or DVD video. Thanks.
Does theia model upscale and create the same quality as the Gaia hq model yet? because so far Gaia HQ is the best for 90% of things, but theia is much faster, so im waiting for either Gaia to be sped up on mac, or theia to produce similar, the same or even better results than Gaia HQ.
ive just tested it and theia doesnt bring or ‘create’ as much detail as Gaia hq , which is a massive shame for me because it is so much faster, and in fact the kind of visual result is more akin to Gaia CG than Gaia HQ, skin faces and detail seem very smooth and create the same kind of pixelation effect as Gaia CG. Please make Gaia HQ faster for mac or make Theia as good as Gaia HQ.
Through experimentation, I have found that when upscaling DVD’s, Video Enhance AI outputs everything in Rec.2020. If I use an AviSynth script on Video Enhance AI’s output and convert from Rec.2020 to Rec.709, then the colors match the original DVD almost exactly.
EXCEPT that the Video Enhance AI’s output is always a little darker. If I adjust gamma by about .02, then the levels match is near identical (1.00 -> 1.02 on a scale of 0 to 10, with 1.0 being unchanged).
In most cases, the Rec.2020 is actually an improvement, so that’s a good thing. But the gamma levels going down slightly should be fixed IMHO.
I’m only able to select CPU for A.I. Hardware. Is that expected? I’m using the AMD RX 5700 XT as an eGPU under Windows 10 (Mac Bootcamp).
Very nice, but it would be very useful to be able to clear the start and end points, in order to do another set of still frame sequences, or video. Currently sometimes I’m able to set a new begin and end, but more often it does not work.
A button to clear prior settings would be very helpful.