Firstly, thank you for making the changes that were requested here regarding files being over written and for the new AI modes however, you really need to make some form of tutorial for this software now. The more options you add the more confusing it becomes. The new AI models may be able to work wonders, but they may as well be Japanese if they cannot be applied correctly. Trial and error is not practical when each setting must be thoroughly tested to asses desired results and that is very time consuming. GAIA CG is still the best setting for the novice trying to enhance SD footage even if it can be a little soft. The ability to load GAIA CG and THEN adjust the sharpness etc., would be a great idea as the base level takes away a lot of guess work. Finally the screendoor or grid effect is now becoming a real problem. Please look into that as it is beginning to render the entire process futile as the final results look like watching HD footage behind well, a screendoor.
What we need is a GOOD description of those different models. Itâs not enough just to describe them roughly. Give us some example pictures (or videos) to see what they are doing under best circumstances with zoom ins on particular hotspots.
Right now Iâm just trying out all models on every video just to see what sticks the best.
Edit: But besides that critics from my side. Your product rocks 
I agree. Its too cumbersome to test everything manually, there need to be best practices and why sort of explanations. User manual or something.
I was able to replicate almost the GAIA-CG (without the grid pattern) by playing with the THEIA settings.But people donât realize that without this pattern GAIA-CG would look even more plastic.
The hardest challenge seems to find the right settings in the THEIA models and they are variable for each input. The thing is that the preview is just too short to have a better idea whatâs going on. Add an option to customize the preview length PLEASE.
Ive actually found that outputting in mpeg4 reduces the grid effect as opposed to the supposedly superior method of outputting to PNG/TIFF. Of course having the codec upgraded to at least h.264 would be a great improvement. also please bear in mind that some people will be watching their enhanced work on a large screen tv so what may look good on a 27" monitor may not be as impressive on a 40"+ tv screen.
Itâs normal, itâs part of the details lost with lossy codecs.
Yes. Its expected. Compression penalty.
I forgot to ask, what settings did you use to get the faux gaia effect, please?
Canât download the windows version from the link above. AWS says âThe specified key does not exist.â Any alt link available? @bowen.wang
While Im guessing it will be obvious to video profesionals to me it came as a surprise that there is no rule for making all videos look better.
Some setting will work wonders on some videos while totally crippling others, so there cant be a single effect that will make all poor videos look better.
I think it would be great to have some kind of automated way of getting a few seconds of video and present the results that each effect would get, so the user can decide which one is best, like the old âvariationsâ in photoshop.
Never mind, I went to my product collection page and got a copy of v1.3.8.
I have noticed a bug with this version.
I cannot complete a process leaving exactly 2 frames before completion.
I have no idea how this happens.
This applies to not just one particular video but on other videos as well.
The process just stops therefore leaving the audio out which I assume is merged when all frames are converted.
Please if anyone is experiencing the same result, share your experience or solution.
Hopefully this gets noticed by staff and gets patched.
Version 1.3.8 is noticeably slower than the beta 1.3.5 for me. Not sure what changed since but it was a nice speed increase that went away with every following version.
On my Mac it does not use my egpu. It uses only the intel uhd 630. I checked the option prefer egpu.
Donât even remember, I was playing with a lot of them until something was close to the GAIA-CG preview I had in mind. Anyway the settings results will depend hardly on the video input and can not be replicated with some other video.
Anyway I think that the grid pattern is already very reduced in this new version.
One more thing, has anyone tried to play with previews on 50/60 fps inputs ? Itâs almost impossible to play around. Half a second of preview on a loop, itâs almost impossible to play with THEIA settings on that conditions. One more time , add longer previews for God sake.
I tested on multiply videos the Theia Detail A.I , and it looks like a wind blowing some parts of the video, thatâs weird. Details moving.
100% restore details is more blurry and undetailed than 0% recovery details? wtf?
GAIA is ultraslow now. +10 secs for each frame to 4k. 1,1 seconds in new mode.
Right now these flaws I can confirm:
-
Speed decreased by about 30% using Gaia HQ or CG. (Donât know about the other models though)
-
I can confirm dency01âs problem with distortion using Theia Detail. Looks like a heat distortion or something like that in some areas. I mean the description says it can lead to distortion but the weird thing about it is that I donât know why it distorts this particular area bc the source is Hi-Res without any special problems (noise etc.) in that place. Other areas where I would expect distortions are fine.
-
Some stability thing. If I preview a video, in 8 out of 10 times it will produce an error if I want to preview the next one or the same one with different settings and I have to close VEAI and open it again. Didnât have that specific problem with previous version. Maybe a problem with releasing the resources after first preview?
In fact it works in reverse.
GAIA speed is the same for me, and now it loads even more GPU computing around 85% when previously if I remember it was around 65% with a GTX 1080Ti.
I download 1.3.8 and tried it again on the MacPro 2010. First time I tried a few days ago it would not install. I thought I would try again. It did install but as with 1.2.3 it crashes on starting processing. But the log file seems more informative.
Failed to load model of type 1 from location /Applications/Topaz Labs LLC/Topaz Video Enhance AI/Topaz Video Enhance AI.app/Contents/MacOS/models/fgnet4-HiFi-tf.tz
Unable to load model of type Tensorflow from path /Applications/Topaz Labs LLC/Topaz Video Enhance AI/Topaz Video Enhance AI.app/Contents/MacOS/models/fgnet4-HiFi-tf.tz
Model type: âtensorflowâ for scale : 2 failed to load correctly from /Applications/Topaz Labs LLC/Topaz Video Enhance AI/Topaz Video Enhance AI.app/Contents/MacOS/models/fgnet4-HiFi-tf.tz
Frame processor failed to setup
Frame Processor: Destroy
Thoughts on what this means? By the way it has a Radeon RX 590 8 GB but is running 10.14.6
