One more thing. Batch processing is not working correctly. Some files claim to have been processed buy haven’t. Others write out a few frames. Had one use a different file name from another clip in the list. Now going one-by-one to avoid this.
1 Like
I spend some time over the weekend to test the new v1.1.1 release.
Indeed, render speed increased significant (I use a 6GB RAM graphic card).
Great: Audio is now rendering for M4.
Remarkable improvement on perserving faces and other structures.
Also in the new version again the CG render modus works best in comparison to the other both.
On the other hand: in comparison to the versions before sharpness, denoising and deblocking works not as good with the new release. The versions before generated remarkable improvements. Version 1.1.1 generates visible improvements, but not as sharp, deblocked and denoised as before. Often the new Version adds small restless new structures on smooth surfaces.
From my point of view the users needs on the end some manual adjustment settings for decrease/increase sharpness, deblocking and denoising.
Also it seems useful to mark faces or structures with different individual settings.
An other idea is, to call the modus, if people are in the footage or not. If no people are in place, the CG Modus from the old version (before v1.1.1) should be offered in the settings…
With v. 1.1.1 comes also the modus, to deblock and denoise without upscaling. Great idea! After a lot of small tests I didn’t recognized remarkable improvements. Perhaps not the best modus in order to improve low quality footage.
The Windows’ Task manager does not reflect a real GPU usage. For more accurate figures one should use 3rd party GPU monitoring software like MSI Afterburner or GPU-Z (both free). In my case with RTX 2080 Ti when VE AI processes video the Taskmanager shows about 25% GPU load while both programs I’ve mentioned show about 70%.
1 Like
Trimming of clips within the app before converting is completely messed up. Perhaps only when doing a batch with several videos. I don’t mean the trim is off by a few frames, either. I mean a 30-second commercial ends up with 4 seconds of video with sound from a different section.
APPLICATION CRASHES
I get an instant Crash of Video Enhance AI v1.1.1 when selecting either Preview or Process of a small mp4 video clip.
A crash report pops up with a send to Apple option. The reason for the crash in the report is:
Termination Reason: DYLD, [0x4] Symbol missing
My platform is iMac (Retina 5K, 27-inch, 2017); 4.2 GHz Intel Core i7; 24 GB 2400 MHz DDR4; Radeon Pro 580 8192 MB; High Sierra 10.13.6.
The application is supposed to work in this environment but does not use the graphic card.
2 Likes
Please raise a support request at the Topaz Labs website.
Yes! I have an iMac 5k 2019 with 4.2GHz quad-core 7th-generation Intel Core i7 processor, 16GB 2400MHz DDR4, 1TB SSD, Radeon Pro 580 with 8GB video memory, and it crashes for me too every single time I try to process or preview a file, even a small one.
I don’t know what that error message would mean that you got (I assume mine said the same thing), but hopefully they can figure it out and create a new update fairly quickly.
I did send a support ticket out explaining my problem, but right now they just said that they sent it on to their technical support to look at.
Currently testing different video and settings. For me the best setting by far is the CGI which cleans up the image beautifully (if a little too much smoothing). I’ve tried both the HQ and LQ settings and there is an improvement, but it does seems to add more noise. Maybe a setting somewhere between the smoothness of the CGI and the detail of the other settings would be good? (or do I have to run the
I would also really like to see an inbuilt de-interlacer. At the moment I am de-interlacing (so my 25i is becoming 50p) but this must be adding noise. I’ve thought about halving the size of the video at the de-interlacing stage (from 576i to 288p) would this make a difference as I have seen comments about pixel and image density)? Once I have a few samples to compare I’ll report back.
Finally should I be running the Denoise/Deblock first or is this inbuilt into the upscale presets?
Both Versions of video enhance AI could Use the Features in The Other AI products For Example Denoise AI.
Mac pro 5.1, macOS High Sierra, Nvidia 1060 6g, don´t work. Message error before installation:
There has been an error.
Your CPU architecture is not supported at this time. The installer will now exit.
Any ideas? Thanks.
Not being a Mac person but the first post says only this:
- Mac Support . HQ/HQ-CG: OV iGPU/CPU support; LQ: TF-CPU AVX2+MKL support
Is your CPU a TF-CPU? i.e. a CPU that supports TensorFlow?
I could not get the app to run on High Seirra at all (it crashes immediately when dragging videos in). I believe it only works on latest iOS (Mojave?). Because of this I’m sadly not running the upscales on my fastest machine (I had so many problems with Mojave that I downgraded my main machine).
I´dont know. But over Parallels desktop works well. Only over Cpu.
When will AMD support and eGPU support be available for the Mac version?
2 Likes
You need to ask support at the main website.
It works well with my Mac mini 2019 i7 and was os Mojave.
So much disappointed it doesn’t work with my egpu (Vega 56), aah hope it will come with updates.
Works well but takes some much time!!
1 Like
AI image enhancing is actually VERY resource demanding. From what I know of this stuff, as of making if faster without making less of a image enhancer, considering the AI algorithm is implemented right, there’s little to be made here since they use [Someone can correct me if I’m wrong on this] FP32, so they are using your GPU raw processing power. Don’t believe those numbers in the Task Manager, mine shows less than 20% usage (on a RTX 2070), but in HardwareMonitor it shows 100% usage and the card runs hot and fans @ full speed when the app is running a task. If they made a module in FP16 that could call tensor cores on RTX Cards then it could probably made much faster, but that would only be available on RTX-class cards (so only from 2060 to Titan RTX, and the Titan V). I guess in the near future with more AI-specific harware being available, specially in NVIDIA cards they should do a move on that direction
Anyone want to check out a bunch of samples can look at my Vimeo Topaz channel here: Topaz Labs Video Enhance AI on Vimeo
I spent a lot of time preparing files for upscaling and then correcting color, adding grain, etc. post-scaling. Pretty happy with the results. Aside from the snail’s pace conversion, of course. But saving old work into a modern format is definitely worth it, IMO.
1 Like
A few questions for Topaz Labs:
-
Is there a known speed difference between rendering to 8-bit TIFF versus 8-bit PNG?
-
Does outputting to certain multiples make render time more efficient? Like would rendering to 4x be actually faster than rendering to 3.85x, etc?
-
On the Windows version, are there any issues with using a ProRes .mov file that was created on a Mac?
Here are my observations if they’re useful to anyone. I’m a film professional who works in both production and post production on national commercials and theatrical features, so apologies if these are nitpicky…
I started with 60i DV footage that was well-lit and professionally shot.
I first used REvisionFX’s FieldsKit to deinterlace in After Effects. It does a decent (sometimes very good) job, but I’m sure an AI deinterlacer would work significantly better (hint hint Topaz Labs). I loaded the resulting 59.94 fps 720 x 480 ProRes 444 10-bit footage into V.E. AI on a MacBook Pro 2017 running High Sierra.
Output to 500% (slightly bigger than 4k, 2400 px high).
Overall, the results are incredible. Wow.
LQ mode (Topaz’s recommendation for SD footage like this) renders looked great but still somewhat had the source footage’s “ringing” from in-camera sharpening. Color ghosting (inherent in all 4:1:1 DV footage) was still present. Faces were noticeably refined and sharp (more than the rest of the image), though there were some occasional strange artifacts in faces. i assume there is a lot of AI synthesizing going on with faces in particular in LQ mode.
HQ mode renders give outstanding results in everything. “Ringing” and color ghosting were gone. Faces were not as refined and sharp as LQ, but there were no strange face artifacts. HQ mode seems to be particularly good at synthesizing detail in hair.
LQ renders’ colorimetry perfectly match the original source. it’s spot-on.
HQ renders’ colorimetry do not perfectly match the original source. Two issues:
- The gamma seems to be higher/brighter, with an adjustment of 0.9166 fixing it in After Effects (which is also the difference between 2.2 and 2.4 gamma) when all color settings are at “HDTV Rec.709”.
- Even after the gamma adjustment, there seems to be less red in the shadows of skin and faces.
MY RECOMMENDATIONS FOR IMPROVEMENTS / YOUR SALES:
Add an optional AI-powered deinterlacing function within V.E. AI that will create a new frame for every field — this would turn 60i footage into 60 (or 59.94) fps footage. Editors would prefer this so they will have the option of using archival footage as clean 40% slow motion for dramatic effect, or they can convert the 60 fps to 24 fps in the editing software using one of many simple methods.
By adding the built-in deinterlacing option, V.E. AI would be seen as a one-step turnkey solution for optimizing archival standard-definition footage. This would likely make it ubiquitous amongst most networks, documentary producers, and post houses that service documentaries. Spending $299 would be a no-brainer. You would just have to market it to those people as a “one-step turnkey solution.”
Allow for AMD GPU usage (most professional post production is still done on Macs).
Add some kind of control/option for the denoising (that I read it incorporates). It would be nice to use or increase this on HQ renders.
If possible, fix the HQ gamma & colorimetry/red/skin matching issue.
Thank you so much for making V.E. AI, it’s going to help myself and many others a great deal.
5 Likes
I’m using a Mac Pro (2013), 3.5 GHz 6-Core, 32GB ram, 12GB VRAM
Tried to install but go error my CPU architecture is not compatible. Is that true?
Would not install. HOWEVER it’s running in Windows 10 vm using Parallel.
