We have another beta available for your testing. This will be released on Tuesday along with the Iris model, so please let us know of any issues you may have.
Recover original detail disabled for interlaced videos
Recover original detail slider works correctly for proteus v3
Rolled back NVEnc to v12.0.16
Older enhancement presets work correctly
Video should load correctly if default preset is set in preferences
If recover original detail is changed then upscaling no longer resets to 20
Misc updates
Known Issues:
HEVC Main 10 Nvidia GPU encoded videos not playable on VLC 3.x, Use VLC 4.0 from the nightly builds
Videos with mismatched metadata and streams will display incorrect duration
Frame number preview length may shorten on app restart
Recover Original Detail
Purpose of this slider is to retain details from original source if the model’s output is overly smoothed
It adds the source texture details to the model’s output to bring back the original details
Lower value might make the output overly smooth and a higher value will also bring in source noise, compression artifacts, etc.
Be careful when adding source noise as this might also show up on output if recover original detail is used
Please look at the tooltip and let us know if the functionality of the slider is understandable from the name and the tooltip
Enhancement model - Iris
Primary aim of the model is to improve human faces
This model is also designed to enhance low- to medium-quality video, particularly, with noise and compression artifacts. The model tends to lose details on high-quality videos
This model is trained to work on all three categories - progressive, interlaced, interlaced progressive - but the main target is progressive videos
Since the model has significant differences from the Proteus model in terms of aim, scope, and performance, we have decided to give it a completely new name
Please test the model and give us your opinion.
Thank you for testing
Please upload problem videos and logs here: Submit Files .
I’m sad about this… I’ve done some tests on interlaced video with Iris model + original detail recovery… and got very good results.
I hope this problem will be fixed with the next releases.
Most likely they put exaggerated scan lines on the ‘TV footage’ to make it look like TV, in this case video AI is just enhancing them even more, I wouldn’t expect different results.
Still unable to get any preview frames when doing actual exports
The temp file is 0 byte throughout the whole process and I only got the first 2 minutes when the whole video was done processing
As much as I love Iris, there is too much dehalo, even when it’s set to zero in manual settings. This seems to be more problematic with some videos, and less so with others.
It seems to be all NVENC output that won’t play on VLC - both the built in Nvidia H265 output profiles, as well as the H264 one, and the custom H265 one I have been using ever since custom encoders were enabled - all produce output unplayable on VLC.
With the H264 one the first 2 or 3 seconds of video never plays or looks corrupted. With H265 You see maybe 2 or 3 random frames, while the audio continues to play.
I tried the nightly build of VLC 4 and it wasn’t a stellar result, the video barely played but was laggy, looked corrupted and audio was not synced, so not really a solution.
If this can’t be fixed in time for the big release tomorrow then hopefully it can by next week as VLC is my go-to media player and there’s no point in producing output that may or not play in the future.
I think @gregory.maddra aka Bloxgate the ffmpeg build master may need to put eyes on this as well
All previous reported issues from the last Beta still present. The morphing of the images still makes this unusable for anything than strict testing.
On a side note, this is the first time I now have to sideload multiple versions of Topaz V3. However, even with copying and relocating the model folder, the separately loaded version of V3 won’t process anything and just errors out requesting contact with support. Can you explain how I side load now with these newer versions? Currently I have to just constantly uninstall and reinstall the older version to actually use it.
Unfortunately the exact same error issue at the end of an encode is occurring on this Mac release as before. I’ve just uploaded the logs for you to look at.
This version did also crash on first install and run (it tried to load the Iris model a few times and then crashed when it started the encode), but I hadn’t rebooted the system first. Upon reboot it ran fine up until the very end when it errored.
I have tried both betas out on my Windows laptop (which have run fine) and I have to say for the footage I’m looking to upscale, the Iris profile does an amazing job now, not just on faces but on guitars as well.
Would be useful to re-introduce the original detail recovery for interlaced footage as the thousands of files of SD footage I am looking to upscale to 4K is all interlaced and I’m sure some of it might need a tweak, so having that option again would be appreciated!
At the moment our suggestion for this issue is to use VLC 4, until either FFmpeg or VLC have a proper fix available.
I haven’t seen the issues you’re still having in VLC 4 with files that I’ve personally been able to reproduce the playback freezes with. Have you submitted your affected files to the dropbox? It’d be appreciated if you could.
I have observed an odd behaviour with this beta. ffmpeg.exe eats up lots of memory just when video file loads. This only happens for files that come from an DVD in NTSC format.
The memory usage crawls up until the system crashes.
This happens on different systems:
Threadripper 3970X with 128 Gb RAM and RTX 2080 SUPER
Ryzen 9 5900 X with 32 Gb RAM and RTX 3070Ti
Both running Win 11 and the latest studio drivers.
Strange is that every other content than NTSC doesn’t show this behaviour.
