so where's all the Topaz Video Enhance AI discussion?

[rrats]

Sorry, this thread isn't really about Hybrid.

Sure it is.

- New post asks question about crappy software. (Topaz)
- Thread eventually starts to discuss better software than can, may, or will do better. (Hybrid)

Common occurrence in many forums.

I know topaz isn't an amazing piece of software, every answer of yours in this thread is straight up shilling. You remind me of linux users who constantly tell me to stop using windows. Recommending somebody hybrid and saving them 200 bucks is fine, but constantly calling it bollocks is a bit too much

[JoelHruska]

Selur,

Here is the SFE-Complete clip I uploaded previously, run through two different Topaz Models. All of my comments below regarding "best" output or the way various models treat content should be understood to refer only to Star Trek Deep Space Nine and Star Trek Voyager. Typically, content from these shows is always run through AviSynth or VapourSynth before upscaling to perform an IVTC + additional processing. I have not done that here.

Gaia-CG (Topaz 1.5.3): This is an old version of Topaz from last September. It performs poorly on the M2V file if the output is not processed first. If given appropriately processed material, this upscaling model produces the best output in CGI scenes of anything Topaz has produced. This file was not pre-processed, so noise is immediately visible on the lower saucer section of the foreground ship.

https://1drv.ms/u/s!AphTLFRW13WMkECkQVNglzOBd208?e=XWnXce

Artemis High Quality (Topaz 2.3.0): Artemis-HQ is the rendering model I recommended readers use as part of my upscaling articles on Deep Space Nine. It is not as sharp as Gaia-CG, but it also less prone to leaving odd noise in a scene. It doesn't offer the best achievable output, but it also doesn't break anything. It introduces very few problems and the problems it causes tend to be subtle.

https://1drv.ms/u/s!AphTLFRW13WMkD83aud_2u7RClIB?e=EFriR9

Here's a fog comparison:

https://imgsli.com/NjM2MzM

Artemis HQ (left) does a much better job than Gaia-CG (right).

You will notice deinterlacing errors in both of these upscaled M2V files because those errors are present and visible in the source DVD. One of the reasons I started using AviSynth was to repair those errors. Because Deep Space Nine and Voyager are both variable frame rate shows, Topaz does not understand the frame rate to set when one loads the entire M2V or MakeMKV-derived file into the application. It sets a default 28 fps frame rate. In order to avoid this across an entire episode, the episodes must be converted to CFR. AviSynth (or VapourSynth) is essential to this process. Topaz cannot perform an IVTC. It has a model, Dione, which can convert a 29.97 fps M2V to 60 fps, but Dione makes the stars flicker and it causes deinterlacing artifacts. They're actively working to improve it, but it can't be used yet.

I have also included my own tuned output below for the same clip.

https://1drv.ms/u/s!AphTLFRW13WMkEGDI2_9mRSZFFkh?e=gbnsn1

This file is a combination of pre-processed Gaia-CG 1.5.3 and pre-processed Artemis High Quality, with the AHQ footage overlaid on top of the GCG footage and blended together. The combined video was then run through Topaz's strong DeHalo filter at 100% size (meaning, no upscale) for just a bit of final post-processing. I realize you did not ask for this footage, but I wanted to illustrate how much better Topaz output is if appropriately pre-processed and assembled.

My current workflow does not perfectly preserve all particle effects and it slightly darkens background stars, but it yields better, sharper ship details with less aliasing and noise. It also repairs the multi-frame interlacing error just after the destroyed starship spirals towards the camera. The net positives across an entire episode substantially outweigh the negatives.

The black flash you'll see on the bottom of the saucer section in all clips when the ship is pierced by phaser fire is baked into the NTSC source. It is also present on the PAL version of the DVD. It's a single frame error that can be fixed in Photoshop or a video editing application as desired.

[DruidCtba]

@lollo: ran your clip through vs-ffdnet (https://github.com/HolyWu/vs-ffdnet) with strengh=10 + Spotless, a bit smooth, but this might be a good filter for this. Also added a video where I added CAS at the end.

Cu Selur

Amigo Selur um ótimo trabalho, a qualidade e nitidez melhoraram muito, parabéns .

Att.

Druid®.

[Selur]

@JoelHruska: thanks

[DruidCtba]

Selur,

Here is the SFE-Complete clip I uploaded previously, run through two different Topaz Models. All of my comments below regarding "best" output or the way various models treat content should be understood to refer only to Star Trek Deep Space Nine and Star Trek Voyager. Typically, content from these shows is always run through AviSynth or VapourSynth before upscaling to perform an IVTC + additional processing. I have not done that here.

Gaia-CG (Topaz 1.5.3): This is an old version of Topaz from last September. It performs poorly on the M2V file if the output is not processed first. If given appropriately processed material, this upscaling model produces the best output in CGI scenes of anything Topaz has produced. This file was not pre-processed, so noise is immediately visible on the lower saucer section of the foreground ship.

https://1drv.ms/u/s!AphTLFRW13WMkECkQVNglzOBd208?e=XWnXce

Artemis High Quality (Topaz 2.3.0): Artemis-HQ is the rendering model I recommended readers use as part of my upscaling articles on Deep Space Nine. It is not as sharp as Gaia-CG, but it also less prone to leaving odd noise in a scene. It doesn't offer the best achievable output, but it also doesn't break anything. It introduces very few problems and the problems it causes tend to be subtle.

https://1drv.ms/u/s!AphTLFRW13WMkD83aud_2u7RClIB?e=EFriR9

Here's a fog comparison:

https://imgsli.com/NjM2MzM

Artemis HQ (left) does a much better job than Gaia-CG (right).

You will notice deinterlacing errors in both of these upscaled M2V files because those errors are present and visible in the source DVD. One of the reasons I started using AviSynth was to repair those errors. Because Deep Space Nine and Voyager are both variable frame rate shows, Topaz does not understand the frame rate to set when one loads the entire M2V or MakeMKV-derived file into the application. It sets a default 28 fps frame rate. In order to avoid this across an entire episode, the episodes must be converted to CFR. AviSynth (or VapourSynth) is essential to this process. Topaz cannot perform an IVTC. It has a model, Dione, which can convert a 29.97 fps M2V to 60 fps, but Dione makes the stars flicker and it causes deinterlacing artifacts. They're actively working to improve it, but it can't be used yet.

I have also included my own tuned output below for the same clip.

https://1drv.ms/u/s!AphTLFRW13WMkEGDI2_9mRSZFFkh?e=gbnsn1

This file is a combination of pre-processed Gaia-CG 1.5.3 and pre-processed Artemis High Quality, with the AHQ footage overlaid on top of the GCG footage and blended together. The combined video was then run through Topaz's strong DeHalo filter at 100% size (meaning, no upscale) for just a bit of final post-processing. I realize you did not ask for this footage, but I wanted to illustrate how much better Topaz output is if appropriately pre-processed and assembled.

My current workflow does not perfectly preserve all particle effects and it slightly darkens background stars, but it yields better, sharper ship details with less aliasing and noise. It also repairs the multi-frame interlacing error just after the destroyed starship spirals towards the camera. The net positives across an entire episode substantially outweigh the negatives.

The black flash you'll see on the bottom of the saucer section in all clips when the ship is pierced by phaser fire is baked into the NTSC source. It is also present on the PAL version of the DVD. It's a single frame error that can be fixed in Photoshop or a video editing application as desired.

Friend JoelHruska, what do you suggest to me to get the best out of VEAI using pre-processing with AVISYNTH or VAPOURAVISYNTH, in this example, where I still couldn't find a good preview of the Chinese audience's faces at the 1999 Yanni's Tribute show, late 1999 part of this show in this video (https://drive.google.com/file/d/1BaXg_ZCDU8tjk5R6F4gVHXV4BqlgGP7-/view?usp=sharing).

Note that this video is 480i, and the closest thing I wanted was with the DIONE INTERLACED ROBUST V4, but it created in some points a SHAKE effect in the image in some points of the entire video (here the original 480i videos and the 1080p results in VEAI 2.2.0 with DIR V4: https://drive.google.com/drive/folders/1UTV3CMdZ7sR8GpOzLlEJNmZsIuJaYs9V?usp=sharing)

And here is a test I did using HYBRID's QTGMC (https://drive.google.com/drive/folders/17v6GxD5ZpQglICgtpDmPTB3r6R8oVbBS?usp=sharing), with the help of my friend SELUR, but the results using the PROTEUS module are not very good as the DIR V4, but they solved the TREBLE problems that you can see in the video 0004 of the link above the tests with the DIR V4.

Below is the video I would like if it is possible for you to analyze for me my friend JoelHruska:

https://drive.google.com/file/d/1BaXg_ZCDU8tjk5R6F4gVHXV4BqlgGP7-/view?usp=sharing

I would like to thank you in advance for your attention and time, if you could please me, because I understand the cost of time spent that many cannot give.

Att.

Druid®.

P.S. My biggest wish is to get a better SHARPNESS of this show in Upscaling 1080p and/or 2160p.

[davexnet]

Any particular reason not to double rate deinterlace? Motion smoothness looks rough in some spots

[rrats]

Any particular reason not to double rate deinterlace? Motion smoothness looks rough in some spots

Because

[davexnet]

Any particular reason not to double rate deinterlace? Motion smoothness looks rough in some spots

Because

That may be so, but the Yanni footage is interlaced not telecined. Whether there's an option in the QTGMC
settings to help with that issue, I'm not sure

[JoelHruska]

Druid,

Sure. Let me look at it.

[JoelHruska]

Druid,

Alright. I've got a couple of script ideas cranking on this. Here's the first one:

Let me know if you like this. I'll also post any of the new output if it's better than this. The clip below maintains 30 fps. Your output is interlaced, not telecined, so I did not attempt an IVTC. I used a combination of QTGMC and TDeint to process this. There's some deliberate noise injected into the video because this helps with Topaz output, but this clip has not been upscaled yet and I've got some different options running with different sharpness and noise levels.

https://1drv.ms/u/s!AphTLFRW13WMkEJsvaNToJUvywRZ?e=aFeMaj

EDIT: Go ahead and take a look at this, but I've already got a tweaked better version with less noise in it. Let me know if this is what you had in mind, though.

[JoelHruska]

Alright. Updated results here.

First, I think this version is better than the one I posted last night. See what you think. It's less noisy. I just tweaked an existing script for it, so this can be fine-tuned to get whatever results you want.

https://1drv.ms/u/s!AphTLFRW13WMkEM7klc4qjgtCKbL?e=1iySQZ

Here's what that same output looks like when put through Artemis-HQ. You mentioned using Proteus and I don't think that would be a problem.

https://1drv.ms/u/s!AphTLFRW13WMkER1CjHiAHfKUPpB?e=K1ruMD

This output is at 30 fps, but 60 fps would not be an issue. If you compare against the clips you posted, you'll find that the version above is less noisy and sharper, as you requested.

[DruidCtba]

Alright. Updated results here.

First, I think this version is better than the one I posted last night. See what you think. It's less noisy. I just tweaked an existing script for it, so this can be fine-tuned to get whatever results you want.

https://1drv.ms/u/s!AphTLFRW13WMkEM7klc4qjgtCKbL?e=1iySQZ

Here's what that same output looks like when put through Artemis-HQ. You mentioned using Proteus and I don't think that would be a problem.

https://1drv.ms/u/s!AphTLFRW13WMkER1CjHiAHfKUPpB?e=K1ruMD

This output is at 30 fps, but 60 fps would not be an issue. If you compare against the clips you posted, you'll find that the version above is less noisy and sharper, as you requested.

Friend JoelHruska, first thank you very much for your time in solving my problem here, thank you very much .

As you can see, I gave two addresses here, one that has the results obtained with 480i sources (originals from DVD ripped to MKV with MAKEMKV) and their respective 1080p Upscales with VEAI's Dione Interlaced Robust V4 module, this one among all the ones that work with DVD Interlaced fonts was what I got close to the result I expected from Topaz's VEAI, which for me was reasonable, considering the font is showy and quite problematic to have a better SHARPNESS in Upscale to 1080p and/or 2160p , because I would never expect magic from Topaz or any other program, even with AVISYNTH and/or VAPOURAVISYNTH, but what also bothered me was the problem of the IMAGE SHAKE in some points of this video show by Yanni, see the video (https:// /drive.google.com/file/d/1Qra9Ymeh9r7rgAMOkUnubhKjJ-Aac2Pk/view?usp=sharing) and it made me very upset.

So you said, at the beginning of the first reply post, that the videos were in progressive mode, yes they are, but because I realized that QTMG does a good job, but the originals, if you need to work on them are here my friend: https://drive.google.com/drive/folders/1UTV3CMdZ7sR8GpOzLlEJNmZsIuJaYs9V?usp=sharing, and the progressives with QTMG are here: https://drive.google.com/drive/folders/1UTV3CMdZ7sR8GpOzLlIEJP que fois ques? you used it with your great work posted here.

Well, here's the considerations and doubts: I remember that in the previous posts you advocate that, based on your tests, the use of grains with avisynth makes you get a better result in the SHARPNESS of the VEAI upscaling, because you used the one with the least grains (the second) or as you said smoother grains?

I want to say beforehand that you achieved a very, very good result in Upscaling, I believe that better than VEAI's DIR V4 module, it was very good and for me I can't see how to improve even more, if that's possible, however I will run some tests here to see the results with other VEAI modules for progressive DVD sources.

Because you used Artemis-HQ and not Artemis-MQ, wouldn't this be better for 480p DVD source videos, or would the result be more plastic, unnatural faces and what I call gouache ink?

You, my friend JoelHruska, are doing a very important job, because your data gives very interesting information to unite the two worlds, where you found that the grains make the VEAI upscaling have a better result, than treating and smoothing the images , because I also realized through some tests here with this show by Yanni - Tribute both using the 480p videos treated by avisynth and QTGMC and the ones that friend Selur passed here using Upscaling with NNEDI3, because the VEAI practically does not improve the SHARPNESS in nothing in the final VEAI product, such as scaling from 480p to 720p with NNEDI3 and then scaling the result video to 1080p or 2160p in VEAI, you noticed the insight that nobody saw and I thank you so much for that .

Well if you are not going to do other tests, I would like to ask you to go through all the step-by-step steps you used here to obtain this result, which for me is very satisfactory, since it solved the problem of the faces of the Chinese audience, which they were very natural and sharper as I expected.

Friend JoelHruska, once again my thanks for your help, and I would also like to thank everyone who came here and posted their opinions, and especially the great master Selur who has helped us a lot here in this forum/topic.

Att.

Druid®.

PS If you think you can do better using the original 480i videos from the show, the address I posted above, to use other parameters from Hybrid's QTGMC, then feel free, I just used the settings that grandmaster Selur gave us here about use QTGMC for deinterlacing .

[DruidCtba]

Friend JoelHruska, a question filled my head at this point after posting my previous answer.

Considering your perception about grain, if using an avisynth or even vapouravisynth script increasing the grain and not smoothing it and using an Upscaling with NNEDI3 and then processing the result in VEAI with Artemis or Proteus module, would the result not bring better SHARPNESS in the final result for 1080p and/or 2160p?

It's just a question that arose here, as the results obtained with Upscaling with NNEDI3 and avisynth smoothing the video didn't improve with VEAI in my tests.

Att.

Druid®.

[DruidCtba]

Another thing you said friend JoelHruska, in answer to my question, about going from 30 to 60 fps, well I don't think it's necessary, and the final result doubles the file size and I didn't like it very much, I prefer it at 30 fps.

There is a lot of controversy about smoothing 60fps images over 30fps, and I would like to say that I researched about this a long time ago, and the interesting thing is that for some people 30 fps gives shaky and 60 fps doesn't or is more smooth, and what people don't know is that this is a personal problem, that is, there are people who retain the images on the retina for longer and others for less time, so some people don't understand why for some this is a problem, but yes a real problem, and for others it is not.

In my case, I retain more images and that doesn't cause a problem for me, and I don't remember the site of this discussion anymore, it even had a video that showed you if you retained more or less images on the retina, so for these fold the fps will be better.

The discussion was about 60fps and 120fps monitors, that's all I remember, so people who retain retinal images for a short time, 60fps TVs are problematic for them and they should opt for 120fps TVs and /or 240 fps.

Att.

Druid®.

[poisondeathray]

about going from 30 to 60 fps, well I don't think it's necessary, and the final result doubles the file size and I didn't like it very much, I prefer it at 30 fps.

It does not double the filesize if you use temporal compression . Only I-frame formats are roughly doubled in size (such as image sequences, uncompressed video, etc..). On average it might be ~1.3x - 1.5x the size for 59.94p vs. 29.97p when using a final distribution format for a similar level of image quality

But you were complaining about "shake motion" Taj Mahal, horizontal, vertical, diagonal - that's mainly because of using 29.97p instead of 59.94p. You discarded 1/2 the data. When you see the original DVD on a TV, you're seeing 59.94 different images / second , not 29.97

Part 003: in this part there is a shake in the Taj Mahal footage in the vertical downward movement that does not occur in the original DVD image, an effect that is also noticeable in some parts of the show in all movements: horizontal, vertical or diagonal .

Part 004: This part shows a more gross problem over the one reported above, from part 003, which is extreme image shake, to diagonal camera movement.

Clips 3 & 4 did not have the other issues 1 & 2 had; clips 1 & 2 have additional issues with duplicate frames, dropped frames, bad scene changes . So this implies the "shake motion" is purely from the 29.97 frame rate issue

[DruidCtba]

about going from 30 to 60 fps, well I don't think it's necessary, and the final result doubles the file size and I didn't like it very much, I prefer it at 30 fps.

It does not double the filesize if you use temporal compression . Only I-frame formats are roughly doubled in size (such as image sequences, uncompressed video, etc..). On average it might be ~1.3x - 1.5x the size for 59.94p vs. 29.97p when using a final distribution format for a similar level of image quality

But you were complaining about "shake motion" Taj Mahal, horizontal, vertical, diagonal - that's mainly because of using 29.97p instead of 59.94p. You discarded 1/2 the data. When you see the original DVD on a TV, you're seeing 59.94 different images / second , not 29.97

Part 003: in this part there is a shake in the Taj Mahal footage in the vertical downward movement that does not occur in the original DVD image, an effect that is also noticeable in some parts of the show in all movements: horizontal, vertical or diagonal .

Part 004: This part shows a more gross problem over the one reported above, from part 003, which is extreme image shake, to diagonal camera movement.

Clips 3 & 4 did not have the other issues 1 & 2 had; clips 1 & 2 have additional issues with duplicate frames, dropped frames, bad scene changes . So this implies the "shake motion" is purely from the 29.97 frame rate issue

Hi friend poisondeathray, what I meant is that SHAKE is much more noticeable in video 4, and a little bit in video 3, but what I'm referring to is the Upscaling of the Dione Interlaced Robust V4 module, you can see that from the link that I posted here above that it's on google driver, and it doesn't happen with the Proteus module, so the problem is being generated by the DIR V4 module, but feel free to prove to me that if I doubling from 30 to 60 fps with avisynth or vapouravisynth I won't get this effect in the final product that VEAI produced as in the DIR V4 module, and this occurs in various parts of Yanni's concert - Tribute.

About me saying it's doubling the size, well this is happening in Topaz's VEAI (some modules that double fps are also doubling the size of the Upscaling final product), and not with QTGMC on Hybrid OK!, not in the range of 1.3 and/or 1.5.

Att.

Druid®.

[JoelHruska]

Druid,

I'll have to answer you in more detail tomorrow, but here's the script to produce the output I uploaded. I copied this over from some of my restoration work on DS9 and VOY -- that's why it's tuned the way it is -- but I tested a simplified version in which I stripped out most of the QTGMC commands and went with default "Very Slow" presets. The final output was not as good -- noisier and less sharp on the whole. I should note that I borrowed part of this from another individual, but it works fairly well.


TDeint()
QTGMC2=QTGMC(Preset="Very Slow", InputType=2, SourceMatch=3, Lossless=2, MatchEnhance=1.0, MatchPreset="Very Slow", MatchPreset2="Very Slow", sharpness=0.4, SMode=2, Rep0=11, Rep1=9, Rep2=9, RepChroma=True, Sbb=0, NoiseProcess=1, ChromaNoise=True, DenoiseMC=True, NoiseTR=2, GrainRestore=0.5, NoiseRestore=0.5, NoiseDeint="Generate", StabilizeNoise=True)
QTGMC3=QTGMC(Preset="Very Slow", InputType=3, SourceMatch=3, Lossless=2, MatchEnhance=1.0, MatchPreset="Very Slow", MatchPreset2="Very Slow", sharpness=0.4, SMode=2, Rep0=11, Rep1=9, Rep2=9, RepChroma=True, Sbb=0, NoiseProcess=1, ChromaNoise=True, DenoiseMC=True, NoiseTR=2, GrainRestore=0.5, NoiseRestore=0.5, NoiseDeint="Generate", StabilizeNoise=True)
Repair (QTGMC2, QTGMC3, 9)
pSharpen(strength=50, threshold=90, ss_x=4.0, ss_y=4.0)
MAA2(mask=1, chroma=true, ss=4.0, aa=128, aac=128, threads=8, show=0)
GradFun3(smode=2)

[poisondeathray]

Hi friend poisondeathray, what I meant is that SHAKE is much more noticeable in video 4, and a little bit in video 3, but what I'm referring to is the Upscaling of the Dione Interlaced Robust V4 module, you can see that from the link that I posted here above that it's on google driver, and it doesn't happen with the Proteus module, so the problem is being generated by the DIR V4 module, but feel free to prove to me that if I doubling from 30 to 60 fps with avisynth or vapouravisynth I won't get this effect in the final product that VEAI produced as in the DIR V4 module, and this occurs in various parts of Yanni's concert - Tribute.

To be clear, I'm replying to and referring to the ones linked to in the original post
https://forum.videohelp.com/threads/399360-so-where-s-all-the-Topaz-Video-Enhance-AI-d...e6#post2624036
https://drive.google.com/drive/folders/1ZopA3bVBQHsiG1bzuj7Fv-uhgDd974Q1

003 and 004
Yanni - Tribute (1997 480i DVD to 1080p TVEAI 1440x1080 by Druid 20121)_003.mkv
Yanni - Tribute (1997 480i DVD to 1080p TVEAI 1440x1080 by Druid 20121)_004.mkv


003 does have duplicates at the beginning causing jerky playback; but the Taj Mahal "vertical downward movement " at the middle and end which you were referring to does not. The only problem in that section is 29.97.

Same with 004. The diagonal camera movement that you indicated has problems, is purely 29.97 (there aren't issues with duplicates frames like the other clips and sections, which are from improper processing)

So that suggests your problem is discarding 1/2 the data. Ie. using 29.97 instead of 59.94

You can go frame by frame to verify this. Or describe more clearly what you mean by "shake" - maybe there is some communication issue

You're complaining about motion issues - but discarding 1/2 the motion data. It's counter productive

[poisondeathray]

what I meant is that SHAKE is much more noticeable in video 4, and a little bit in video 3, but what I'm referring to is the Upscaling of the Dione Interlaced Robust V4 module, you can see that from the link that I posted here above that it's on google driver, and it doesn't happen with the Proteus module, so the problem is being generated by the DIR V4 module, but feel free to prove to me that if I doubling from 30 to 60 fps with avisynth or vapouravisynth I won't get this effect in the final product that VEAI produced as in the DIR V4 module, and this occurs in various parts of Yanni's concert - Tribute.

Or describe more clearly what you mean by "shake" - maybe there is some communication issue

Ok, comparing QTGMC and Proteus output at 29.97 - "Shake" is probably not an ideal term to describe this

"Shake" implies some motion, perhaps some x,y translation, or rotation . e.g. camera "shake" would imply frame moving up/down/left/right in some irregular pattern

I think you're describing the deinterlacing artifacts and frame to frame inconsistencies from bad deinterlacing and perhaps single frame upscaling - objects are not temporally consistent and change shape, flutter from frame to frame.

[JoelHruska]

Druid,

Dione is not currently functional for most content and cannot be relied upon. This is unfortunate, because it actually does a great job preserving background detail in my own project in a way I would like to hold on to. Unfortunately, there is one way and one way only to use Dione right now.

1). Pre-process a clip in AviSynth and upscale it in Artemis-HQ or whatever alternate model you intend to use. We'll say the final output here is 23.976 but it doesn't matter if it's 29.97.
2). Process the same clip through Dione. The end result here is a 60 fps file with a lot of deinterlacing problems.
3). Compare the Dione frames with the Artemis-HQ (or whichever model) frames. Find the 60 fps frames that match the 23.976 fps frames identically. Merge the frames together and blend the output in a way that lets the benefits of Dione through (I typically have blended it at around 33 - 50% opacity).

Alternately, use masking and an application like DaVinci Resolve to combine the footage together *after* you've hand-combed the Dione output for the proper frames to save. This is probably possible in AviSynth too, but I do not know how to do it in that application.

This is absolutely possible but it's also a slow, painstaking process. It's the sort of thing you might possibly do for a few dozen frames or to patch up a particular bad spot in a video, but it's not reasonable for any length of footage.

I'll give you another example of how Dione breaks. I have a project I'm working on where Dione producs beautiful output, *except*, it reverses every 60th frame. In every case, frame 60, 120, 180, 240, etc was swapped with frame 61, 121, 181, 241, etc. I had to swap the file names around to correct the frame order myself.

Do not trust Dione. Do not expect functional output from Dione. If you *get* functional output, feel very surprised and lucky. I am hopeful they'll keep working on it, but it's not there yet.

[DruidCtba]

[QUOTE=poisondeathray;2627741]

You can go frame by frame to verify this. Or describe more clearly what you mean by "shake" - maybe there is some communication issue

You're complaining about motion issues - but discarding 1/2 the motion data. It's counter productive

Hi friend poisondeathray, I don't know if you have VEAI with you to do tests, so let me again try to clarify something that I thought was already clear to you, the problem is not in the fluidity you think you get with 60 fps, if this if that were the case, the result obtained by the Proteus module would also have this problem with 30 fps, since you insist that this is the problem with TREMIDAS, and no, when I use Proteus it generates a result at 30 fps and TREMIDAS DOESME, that's it that I'm trying to tell you, one thing is theory and another thing is practice, finally with the help of my friend JoelHruska I already have a guide on how to solve the problem of plastic, unnatural faces, gouache painting effect that other VEAI modules generate, and that DIR V4 solves the best possible way, so let's not go into this discussion any further, because I'm now close to getting through avisynth/QTGMC/VEAI what I've been looking for all these days, and thanks to everyone here who helped in private friend Joe lHruska and the master Selur .

Att.

Druid®.

[DruidCtba]

Druid,

Dione is not currently functional for most content and cannot be relied upon. This is unfortunate, because it actually does a great job preserving background detail in my own project in a way I would like to hold on to. Unfortunately, there is one way and one way only to use Dione right now.

1). Pre-process a clip in AviSynth and upscale it in Artemis-HQ or whatever alternate model you intend to use. We'll say the final output here is 23.976 but it doesn't matter if it's 29.97.
2). Process the same clip through Dione. The end result here is a 60 fps file with a lot of deinterlacing problems.
3). Compare the Dione frames with the Artemis-HQ (or whichever model) frames. Find the 60 fps frames that match the 23.976 fps frames identically. Merge the frames together and blend the output in a way that lets the benefits of Dione through (I typically have blended it at around 33 - 50% opacity).

Alternately, use masking and an application like DaVinci Resolve to combine the footage together *after* you've hand-combed the Dione output for the proper frames to save. This is probably possible in AviSynth too, but I do not know how to do it in that application.

This is absolutely possible but it's also a slow, painstaking process. It's the sort of thing you might possibly do for a few dozen frames or to patch up a particular bad spot in a video, but it's not reasonable for any length of footage.

I'll give you another example of how Dione breaks. I have a project I'm working on where Dione producs beautiful output, *except*, it reverses every 60th frame. In every case, frame 60, 120, 180, 240, etc was swapped with frame 61, 121, 181, 241, etc. I had to swap the file names around to correct the frame order myself.

Do not trust Dione. Do not expect functional output from Dione. If you *get* functional output, feel very surprised and lucky. I am hopeful they'll keep working on it, but it's not there yet.

Perfect my friend JoelHruska, Dione is really unreliable, where he hits and most of the other VEAI modules fail, and it was the thing that upset me the most, when finishing Yanni's Upscaling - Tribute after 50 hours, he it also generates unacceptable glitches, so much so that the current version doesn't even support it anymore.

But here's what you suggest and realize that through an avisynth device, which generates graininess in the result of the pre-preparation of the video through avisynth and QTGMC to result in a progressive and grainy video to be used by Gaia, Artemis and others modules that handle progressive files and according to the result you presented here, they improve the SHARPNESS so much, which is what I always looked for, that I no longer need the DIR V4 or other DV model for interlaced videos, and the result was that with this technique of yours it is possible today that I have the same result as DIR V4 for the natural faces of the Chinese audience, as well as faces far more natural than those obtained by the other modules with 480i/576i source or treated before by me with QTGMC just no grain and smoother and using SD progressive font modules in VEAI.

I'm happy with the result, which you got with your studies, and as in all my saga here on this forum, as well as in the official TOPAZ VEAI community, which was to get an Upscaling result with the highest SHARPNESS possible (very weak in Upscaling from my TV Samsung TU8000 50"), in a 1080p and/or 2160p upscaling and with natural faces, which currently aren't quite as close to the quality of the DIR V4 module.

Att.

Druid®.

[DruidCtba]

Meu amigo JoelHruska, eu posso usar este seu script no StaxRip v.2.60?

TDeint()
QTGMC2=QTGMC(Preset="Very Slow", InputType=2, SourceMatch=3, Lossless=2, MatchEnhance=1.0, MatchPreset="Very Slow", MatchPreset2="Very Slow", sharpness=0.4, SMode=2, Rep0=11, Rep1=9, Rep2=9, RepChroma=True, Sbb=0, NoiseProcess=1, ChromaNoise=True, DenoiseMC=True, NoiseTR=2, GrainRestore=0.5, NoiseRestore=0.5, NoiseDeint="Generate", StabilizeNoise=True)
QTGMC3=QTGMC(Preset="Very Slow", InputType=3, SourceMatch=3, Lossless=2, MatchEnhance=1.0, MatchPreset="Very Slow", MatchPreset2="Very Slow", sharpness=0.4, SMode=2, Rep0=11, Rep1=9, Rep2=9, RepChroma=True, Sbb=0, NoiseProcess=1, ChromaNoise=True, DenoiseMC=True, NoiseTR=2, GrainRestore=0.5, NoiseRestore=0.5, NoiseDeint="Generate", StabilizeNoise=True)
Repair (QTGMC2, QTGMC3, 9)
pSharpen(strength=50, threshold=90, ss_x=4.0, ss_y=4.0)
MAA2(mask=1, chroma=true, ss=4.0, aa=128, aac=128, threads=8, show=0)
GradFun3(smode=2)

Att.

Druid®.

P.S. What are the commands that generate the granulations, just out of curiosity .

[JoelHruska]

NoiseRestore and GrainRestore both generate what you call "granulation", but noise helps the upscaler more. You should know that the boosting effect is much weaker than it used to be. Boosting noise to improve image fidelity still works well in Gaia-CG 1.5.3 but not so much in Artemis-HQ 2.3.0. You can expect a small fidelity increase, but do not overdo it.

EDIT: Remove "ChromaNoise=True" if you intend to push NoiseRestore above 0.5. If you keep that line of code and increase NoiseRestore to 0.6 or above, you'll start getting a green tint to your footage.

[DruidCtba]

NoiseRestore and GrainRestore both generate what you call "granulation", but noise helps the upscaler more. You should know that the boosting effect is much weaker than it used to be. Boosting noise to improve image fidelity still works well in Gaia-CG 1.5.3 but not so much in Artemis-HQ 2.3.0. You can expect a small fidelity increase, but do not overdo it.

EDIT: Remove "ChromaNoise=True" if you intend to push NoiseRestore above 0.5. If you keep that line of code and increase NoiseRestore to 0.6 or above, you'll start getting a green tint to your footage.

My friend JoelHruska, do you know how I can install this Gaia-CG module from version 1.5.3 on version 2.3.0?

I ask because when you install a new version, it asks if you want to keep your old modules, it seems strange, but it seems to me that Topaz wants to keep their customers up to date with it, anyway there's how I can install this Gaia-CG version 1.5.3 on my VEAI version 2.3.0?

Att.

Druid®.

P.S. Thanks for the answer, I'll have to do some tests here and see how you behave, friend .

[JoelHruska]

"My friend JoelHruska, do you know how I can install this Gaia-CG module from version 1.5.3 on version 2.3.0"

You can install both simultaneously. I have not figured out a method of getting 2.3.0 or 2.4.0 to recognize 1.5.3's models.

If you install both simultaneously, install 1.5.3 first. Allow it to install to the default directory. Then, install 2.4.0 (just came out yesterday). Install it to C:\Program Files\Topaz Labs LLC\Topaz Video Enhance AI-240\. Create a new desktop shortcut for yourself pointing back to 1.5.3 (Topaz will always point the desktop shortcut to the most recently installed version).

They *should* not interfere with each other.

Be advised that while noising Gaia 1.5.3 produces the sharpest output I've ever seen Topaz produce, this is not always a good thing. There is an interplay between Topaz and AviSynth, and you may need to turn sharpening down in AviSynth before using noised Gaia-CG output in 1.5.3. Also be aware that while 1.5.3 is the sharpest output, it is also sometimes prone to over-amplifying noise.

Two other thing to know: While injecting noise into the video stream up to NoiseRestore=1.0 (remove ChromaNoise=True if you try this!) will sometimes continue to improve sharpness, some of that noise will bleed through into the final video. This is why noise tuning is important when using this method. Also, injecting noise into the background sometimes results in background objects getting sharpened more than foreground objects. How much difference it makes to use noised Gaia-CG 1.5.3 as opposed to Artemis-HQ depends on your source content.

https://imgsli.com/NjQ3Njg

This is the difference between noised Gaia-CG 1.5.3 and Artemis-HQ v11 (from Topaz VEAI 2.3.0) in the Season 3 episode "Defiant" of Deep Space Nine.

https://imgsli.com/NjQ3Njk

Here's the same comparison from Sacrifice of Angels, an episode from Season 6 of DS9.

Sometimes I blend AHQ and GCG output together if I like some of what shows up in both outputs.

[forart.it]

Hi there, I recently found this Upscale "render" that has an interesting workflow:

Flowframes (framerate interpolation): https://nmkd.itch.io/flowframes
Topaz Video AI (resolution upscale): https://www.topazlabs.com/video-enhance-ai
Transition Chooser to clean up scene changes and artifacting: https://www.magentacloud.de/share/lg6gou41fc

Judge result yourself:
https://www.youtube.com/watch?v=iUPVczZ2Xq4

...how to implement it with open source softwares only ?

Hope that helps/inspires.

[rrats]

Hi there, I recently found this Upscale "render" that has an interesting workflow:

Flowframes (framerate interpolation): https://nmkd.itch.io/flowframes
Topaz Video AI (resolution upscale): https://www.topazlabs.com/video-enhance-ai
Transition Chooser to clean up scene changes and artifacting: https://www.magentacloud.de/share/lg6gou41fc

Judge result yourself:
https://www.youtube.com/watch?v=iUPVczZ2Xq4

...how to implement it with open source softwares only ?

Hope that helps/inspires.

VSGAN, any ol' framerate interpolator (like DAIN), Da Vinci resolve

[Selur]

Flowframes -> is open source
Topaz Video AI -> could be replaced with VSGAN+models and Vapoursynth/Avisynth scripts
Transition Chooser -> depending on what you want to do the author might give you the source code (also it is free to use afaik), also using some Vapoursynth/Avisynth script viewer and some knowledge of the respective script language would also allow you to do what it does.

So, I see no real problem other than someone would have to spend tons of time to combine stuff,...

Cu Selur

[DruidCtba]

You can install both simultaneously. I have not figured out a method of getting 2.3.0 or 2.4.0 to recognize 1.5.3's models.

If you install both simultaneously, install 1.5.3 first. Allow it to install to the default directory. Then, install 2.4.0 (just came out yesterday). Install it to C:\Program Files\Topaz Labs LLC\Topaz Video Enhance AI-240\. Create a new desktop shortcut for yourself pointing back to 1.5.3 (Topaz will always point the desktop shortcut to the most recently installed version).

ОК friend JoelHruska, thank you very much for your reply, i'll see how to download version 1.5.3 .

Att.

Druid®.