[If you are posting a new bug, please include the following]
I should have stuck with v3.5.4! No, REALLY! Not joking.
Upscaling Clean PAL 720x576 25fps to 1280x720p 25fps using Proteus:
Presets from v3.5.4 do load in but results are different and bad.
Auto mode looks terrible - denoising is too severe, sharpening is causing extra
‘halo-like’ artefacts that look like old analogue video that’s been over-enhanced,
or has an unterminated cable that’s ringing. Really bad.
Estimate mode is the same as Auto - unuseable.
Denoise is too aggressive even down on setting of ‘2’ and also causing loss of detail
on faces and areas of fine detail that just get flattened.
Details in the original that looked OK are being smeared badly.
Manual mode is unable to do any better. Can certainly make it worse though.
Rendered segments are being deselected and then say “Unprocessed” OMG!
Uninstalling and going back to 3.5.4, sorry guys.
Win 11.23
CPU 12th Gen Core i7 12700H with 64gigs
GPU GeForce RTX4050 6GB Vram
Your log files (Help > Logging > Get Logs for Support) logsForSupport.zip (1.7 MB)
Any screenshots as necessary
[Please be sure to have searched for your bug before posting. Duplicate posts will be removed.]
Do you have the Previous Models option turned on in the Preferences Panel>Advanced Features?
If your presets were using older versions of the AI models this could be where you are seeing the difference in v5.1.4 as it will default to using the latest version of the model.
The devs are working on fixing the preview issue now.
No, I didn’t. I have now and it looks better, thanks.
When I updated from 3.5.4 to 5.1.4 I skipped 4.x.x. The options in the Manual settings for Proteus looked the same as 3.5.4 and frankly I was surprised that the older presets just stayed in the list and loaded in without any warnings so I went with that. Live and learn I guess. Cost me a lot of time and made me feel anti-5.x.x.
I had 3.5.4 installer ready to replace 5.1.4 and only because of your reply I was saved from that ordeal. Thanks again.
Sadly, I’m still having issues. It’s a bit better, but compared to previous conversions I’ve done in earlier VAI versions it’s like I’m wearing old spectacles. Hard to define it more scientifically. Things that are recognisable, e.g. hands, are smeared instead of sharpened, individual hairs that I was amazed to see in conversions I did previously are not happening any more, faces just don’t look right, etc.
I know I’m comparing “apples and oranges”, but VAI’s output just doesn’t look great for my content any more. The sources I’ve used are pretty good and I want to make them look better, not give them vague new problems, so I hope you can get back on track.
Not everybody wants to convert UHD quality stuff up to 16K, I just want to make nice looking, sharp and noise-free, modest upscales of stuff that needs the nice boost I used to get. Filesize is an issue as is the picture quality of 720p content on a, minimum, 65" OLED screen and up.
There’s a Topaz video online called “Video Enhance AI - Using Proteus for fine-tuned video enhancement”. There’s a Cheeta and a sleeping girl in it and it demonstrates how to proceed with the first Proteus - definitely not for 5.1.4’s Proteus!
I have uninstalled 5.1.4 and reverted to 3.5.4 because I’ve wasted a week and need to get some nice results without having to run test after test after test after test after…
Understood, the nice thing about the Video AI license is you can use any previous version at any time and go back and forth between versions when desired.
Ill take a look at that video as I have not seen that one.
I get it, passing along the info and making posts here on the forums it helpful to the devs to see what is going on and what users are running into. They are working to address each of the bugs, issues and concerns that users bring up.
I’m always keen to upgrade to enjoy features that are lacking, but I did not find the new v5.1.4 performance and GUI to be improved relative to v3.5.4, in fact, quite the opposite.
I’d like to see the dev’s make VAI display ALL changes to models and presets in real time on a still frame. If it looks good then a preview could be rendered. The current system is very cumbersome and time-wasting.
If the dev’s can add such a useful facility to VAI perhaps they could approach Nevercenter to incorporate the video side of their CameraBag Pro software (which is available for Mac and Windows) as a Plug-In. CameraBag Pro is nice to use and could also improve PhotoAI as a PlugIn.
I’d like to be able to assess, and fix if need be, the audio quality within the Previews process please. I am used to hearing as well as seeing what I’m doing. Maybe you could also make it so that a capable audio editor like e.g. SoundForge could be used as a PlugIn.
I hope this helps stimulate some interesting and productive improvements.
