Assuming VEAI uses FFMPEG to encode the output, can we expect to see options in the future that give more control? There are competing similar video enhancement products that allow the user to select command line options that are passed to FFMPEG to control the output.
1 Like
I have a Mac Mini with an M1 chip I can use for testing if needed along with a bunch of other Mac/PC hardware.
OK, my PC crash is a 1.7.x version bug. Maybe AMD driver handle, or updated code, I donāt know.
I uninstalled 1.7.1 version, and installed 1.6.1 version. This no include any AMD acceleration, and try with pure CPU encoding. Unbelievable!
No crash, CPU load is 50-60%, memory 27% (5 GB from 20), GPU 9% (with 51 deg. C). Joke.
Can encode without any problem, and have more resources too for other working.
Time is double like with GPU, but more stable.
Here is my settings under 1.6.1 for 3GP video and no problem:
- Theia-Detail (50, 80, 20)
- Scale: 600%
- MPEG4 video
- AI Resources demand: 500%
- AI hardware: CPU
Iāve had stability issues with my AMD RX 480 and found that underclocking a bit, around 10% or 15%, makes it super stable. Itās been running 19hs at the time of writing.
Which software use for underclocking and how?
I underclocked my RX 480 by about 10% and now itās stable (19hs of processing at the time of writing), dropped 10Ā°C and seems to be consuming 105w max, opposed to the 125w I used to see.
It made Gaia-HQ about 0.04 frames/sec slower but I can live with that.
I am having lots of audio issues since purchasing the software. Audio is currently about ~1000% speed up in the output. Any way to fix this? As of now the software is pretty much unuseable due to the audio issuesā¦
Thatās point (and solution) for 1.7.1!
Here is step-by-step solution for AMD cards:
- Open Radeon software (on desktop right click)
- go to Performance tab
- select Tuning tab
- Add Game Profile (on right corner) and find Topaz Video Enhance AI.exe
- Set Tuning Control to Manual
- on left enable GPU tuning, and set Max Frequency (%) to -10
- on right side enable Power Tuning
- set Power Limit (%) to -10
- click Apply Changes and close Radeon Software
Test it!
I get CPU load 80-93% and my card heat is 62 deg. C. while encoding.
I made two longer conversion without crash. I hope this will leave in futureā¦
Update: I was happy too early: while encode third video, I get crash again 
Need more time.
Update 2: I reduced memory clock speed to minimal (1500 MHz now). One video encodedā¦
Update 3: VRAM tuning went to original (disabled), and decreased GPU Max frequency with -10. I updated my guide.
And here is my tests:
- with Gaia models clock speed no goes to maximum, but power consuption went to 65W!
- with Theia Detail and Artemis LQ my GPU clock speed run on maximum (I have now 1113 Mhz for VEAI), but power consuption went to 55 W onlyā¦
You need decrease power and GPU speed with -10 for VEAI and maybe will stable encoding under AMD cards.
1 Like
For your information, I have tested on my old RX580 4GB card, very stable, encoded for few days 7x24, around 75deg max, no under voltage or clock needed. GPU workload set to Compute or Graphics no differences.
Ryzen 1700 (8cores 16threads) CPU overclocked to 3.8GHz and water cooled.
Windows 10 64bit , 32GB ram.
I think a few RX series unstable with default settings, and goes to black screen. I donāt know why. Just encode, and crash. GPU load is 85-95%, and heat is normally 62 deg. C. I donāt know whatās problem with GPU encode. Hasselh tip was great and testing my system with underclock, and I see these steps whats has fixed my crashes.
Topazās ads are quite good at deception. The truth is some features do not work. Audio typically does not work properly. 1000x too fast. Now that is funny! Audio needs to be split from video, audio processing in VEAI disabled, and later remux your audio back into your final video. Do NOT use VEAI to encode a video to .mp4. Avi input and png, tiff output works. Video enhancement works well (with some models and VEIA versions) for 101 to 200% upscale. 400% upscale to 4k and 8kā¦ only if your source is high bitrate, like from Bluray. 100% does nothing. You need to upscale by 200% and resize your output (back to 100% if that was desired).
Are you new to audio-video work? No one can help unless they know your level of experience.
1 Like
after encoding with VEAI, use mkvmerge to mix the VEAI video output with source audio track
Output in png images when encoded with FFMPEG results in correct FPS but longer video, with 20s ca more than input, causing audio sync issues: during the video audio goes out of sync in positive and in negative, maybe duplicated frames? I am converting from 23.96 image sequence to 25fps video, I want to speed up the video, not duplicate or drop frames.
The string of FFMPEG I am using to convert image sequence to mp4 is the following:
.\ffmpeg -framerate 25 -i %06d.png -vcodec libx264 -refs 5 -crf 12 -preset veryfast out.mp4
I gave up on FFMPEG. More reliable is Avisynth with Virtualdub2:
In your .avs file:
ImageSource(āu:\wb\run1%06d.pngā, fps=25, start=000000, end=038460, use_DevIL=true, pixel_type=āRGB24ā)
Change path, fps, and āendā to suit.
Thank you very much! Is it a problem for audio/video sync to start from 23.96 image sequence and go to 25fps video?
Well youāre going to have to change the length of the audio (shorten by 23.976/25). Either resample if pitch doesnāt matter, or do it with pitch correction (3 hour process in Audition with the HQ settings) if pitch is important.
I just did a 25fps to 23.976fps upscale project. I changed the framerate via ImageSource and lengthened the audio in Audition. I noticed the voices in the 25fps release sounded unnaturally high pitched, so I guess they just sped up the film scanner to make it PAL and didnāt do pitch correction. Which meant I only had to resample to the new length (quick), and the pitch went down to the likely original pitch. I used the calculator to determine the new length, but it was off by 100ms at the end of the movie, so I had to redo the audio 100ms longer and remux.
I came across a bug and this seems as good a place as any to put it, but there is a significant and consistent artifact issue when using the Artemis LQ v8 AI model. Artifact may not be the right term, but in all outputs using this AI model there are 4 distinct boxes that appear that are not a part of the source image.
The left image is the source, the right, you can see the 4 boxes.
I love the Artemis models but they both create these artifacts in 1.7.1. Gaia does this to a lesser extent, but it is a general problem with the new way 1.7.1 is processing frames in pieces. The results are amazing otherwise, so we can only hope they offer an option to not process the frames in these sections and avoid the artifacts.
Thanks for the heads up. I am a video editing professional and figured they would get the muxing part right since it had been around for agesā¦
Can you please elaborate on the ādo not use mp4 exportā?
I am currently trying to upscale certain parts from longer videos by selecting a frame range within VEAI but audio is broken 9 out of 10 times. Since I am only converting certain parts simply remuxing afterwards is not an option.
My current workflow therefore is horribleā¦
Extract part I want to upscale into lossless codec (huge filesā¦)
Upscale to proceed (even bigger files)
Reapply audio track in NLE
Export into target codec
Highly annoying!
Iāve also noticed that the software is unable to handle variable frame rates - any workaround here thatās not a total pain in the butt?
I splitted a long video into several smaller parts in order to render it in different times of the day (I use mp4 CRF 0, theoretically lossless), and 2 days later when I joined the files I realized that some frames are missing at the beggining of each part.
I found out that this didnāt happen with AVI files, only MKVs as far as I tested, havenāt tried with MP4 yet.
