Site duplicate rules as applied to AI enhancement
-
@john32123666 I would have agreed with you until I saw the "Origins" series from Falcon. Those studio remasters are appalling and clearly done without any care or skill. I have seen much better remasters of some of those movies made by users on this site, but it's becoming impossible to find the good ones among all the garbage.
-
Re: Site duplicate rules as applied to AI enhancement
IMHO I do not see them as duplicates in general - as the intent is to take source A and make it 'better' to target B with significant alteration of the original. With the AI upscalers they are creating new pixels in an interpretative way based on some trained ML algorithm, some much better than others, and creating re-imaged new content. The point being is that the new AI treated output is different in someway - be it counted in pixel density, resolution, color grading/corrections, vocal isolation/sound restoration, etc.
That said, the problem isn't so much the intent but rather the outcomes of posting poor quality upscaled content. Since the original source isn't destroyed in the process, people shouldn't really care if it were not the fact that downloading large sized dribble is going to take a hit on their ratio as well as wasting time & resources to download it. If ratios weren't a factor it would be just another 'oh well .. delete and move on' moment after downloading a large sized dog turd. Compounding matters, poorly AI-treated content also floods available options when trying to find particular content and trying to pick the good version from the bad. Point here is the problem is not so much duplication, as that's the symptom of the real problem of more dog-water quality content coming in because people think they're contributing.
That problem is getting worse lately - and I fully admit I've contributed to it with some of my earlier attempts. In my case, the constructive feedback I received from the user comments led me to dive much deeper in to the process and not just go with the out-of-the-box approach that Topaz, and other vendors, purport how easy-button it is. Maybe as these fledgling consumer-grade technologies mature to the point where out-of-the-box is generally good most of the time with any content, then reciprocally we will see much more quality content contributions. Or perhaps there will be more that took the route I did and just dive in and learn how to do it better.
However, until such time, assuming such a day comes in the near future if at all, then an interim approach to the problem might be a rule requirement that all AI treated content requires before & after pictures (from the actual source used and output made - not images grabbed from the internet) and some details as to how it was treated. That way we can decide if the content is worth the ratio hit.
-
FWIW, I find that Topaz output is very nice provided that you don't upscale - only do the AI sharpen, noise removal & artifact cleanup at original res & frame rate.
And I know this will make y'all cringe but.... Topaz's file bloat can be handled by Handbrake. Pick a Handbrake quality just high enough to where your casual eye sees no difference on a decent computer monitor, and it will still cut the Topaz bloat by half.
Problem is, one Topaz run takes DAYS on a conventional laptop - so the question asked here, about AI enhances being subject to duplicate rules, is a problem I can't even have, yet.
-
@blablarg18 It all depends on the quality of the source video, and how much useable data there is in the video stream, as opposed to noise. There is a point at which there is too much noise in the video stream for the AI to function as intended, which results in all the Frankenstein faces and porcelain skin we see in most of the uploads on here lately. This will happen regardless of trying to upscale or not.
Upscaling is really only possible with a high quality, minimally compressed, and clean source video. From what I have seen very few of the upscaled videos shared here have come from high quality sources, indeed one was a low res rip from a streaming site, and others were highly compressed with legacy codecs, so any attempt at upscaling will fail. That said, trying to upscale any 480i video to 1080p or higher is just madness, and shows that the user has no understanding of what they are doing. I think some users of the software become disillusioned with the upscaling very quickly because they are either using very poor source videos or aiming way beyond what is possible with the video they are using.
I too use an external program to compress Topaz output, because I don't like how there is virtually no control over the compression in Topaz. It's worth keeping in mind though that you need to export your video from Topaz in a lossless format and not already compressed. Compressing an already compressed video is never recommended because you will begin to add back more of the compression artifacts you have been trying to remove, undoing all the days of processing. I'd recommend using Hybrid (https://www.selur.de/downloads) for compressing, it takes a little time to learn but it's far better than Handbrake, and gives you much more control. BTW If your source was h.264, h.265, AV1 or some other modern codec it's always best compress it with the same codec. Each codec has its own limitations and drawbacks so it's never a good idea to add a new codec with all its issues into the mix.
-
@jrewingwanna said in Site duplicate rules as applied to AI enhancement:
IMHO I do not see them as duplicates in general
I was thinking more about people posting competing upscaled versions of the same video. They would both be 1080p versions and might be similar in size, but could be radically different in video quality. Are they duplicates?
I understand that this isn't a problem yet, and might never be.
-
@john32123666 said in Site duplicate rules as applied to AI enhancement:
Unless the source enhancement/upscaling coming from the studio itself ...
Even then.... I bought the Falcon Vault Collection of their 10 classic vintage films, all of them supposedly carefully remastered. The remastered versions weren't much better, if at all, than the originals. And in one case, Spokes, I could argue that Falcon made it worse. And these weren't even upscaled, just "remastered."
There's a website that is studio-authorized to play and allow downloads of classic videos. I've forgotten the name, but it's similar in nature to the gay-world-dot-org site. They started making 1080p versions of some of their older videos, and I downloaded a couple. They were almost as bad as most of the upscaled versions posted here. Honestly, it looked like they had just stretched the videos out and called it good, without bothering to do any cleanup.
Drekkin can speak more to this than I can, but my understanding is that this kind of remastering/upscaling will only work well if you have a really excellent master video to start from and you hire professionals who know what they are doing (and be prepared to pay for that expertise). Otherwise, you get the kind of junk we see here.
I just wish more of the uploaders here would stop thinking about whether they could do the upscaling and start thinking about whether they should do the upscaling. For me, anyway, for now, the answer is that I shouldn't do it myself, and that it's not worth my time or my bandwidth/ratio to download the work of others.
-
@NF16 There are a few points here I can add something to.
About studios remastering old movies, it's just not going to be cost effective for them in the current market. I mentioned before that the cost of re-digitising old movies, that were originally shot on film, with modern technology would be prohibitively high, considering the tiny amount they would sell. I would also guess that getting a pro to remaster old movies wouldn't be very cost effective either, because there really isn't a market for individual titles anymore. Consumers are used to an "all you can eat" model like Netflix or Spotify, so expecting consumers to pay the price of a subscription for just one old movie doesn't make any business sense these days. It makes far better sense for the studios to put all their old content, in whatever digital format they have, either on their own subscription site or one one of the subscription streaming platforms. That way they can still make a few cents from their old titles without making any effort.
I have seen a few legal sites with 720p and 1080p versions of older movies that would have been shot in SD. One user here shared a load of Pacific Sun movies from one of those sites, and they were all just 480i DVDs ripped at the wrong resolution, and often the wrong aspect ratio and not deinterlaced. Absolute garbage and not worth anyone's money. Imagine a musician releasing a remastered lossless version of their classic 1970s album, and finding out that it was just an old 8-track copied to a PC in 128kbps and then converted to 24-bit FLAC.
I mentioned the Falcon series further up, it's funny you mention it too, because I'm still not over how bad they are, and that Falcon thought they were good enough to release. I actually wouldn't consider them studio remasters, because they haven't used the original recordings. They used DVD rips, and not even high quality rips, because they weren't deinterlaced correctly, so the combing artefacts are now burned into the image forever. How exactly they were "remastered" isn't at all clear, because they just look like they were very crudely processed with a load of Instagram style filters. It's 100% an amateur job, because I don't know any professional who would do that.
You're absolutely right about the quality of the source video for remastering or upscaling. There is a golden rule:
in = out, and that is 100% true. You don't need to be a professional, there are some forums I am on where amateurs are restoring old videos to an amazing quality, but they have been doing it for years, and sharing their failed attempts for advice. Anyone remastering older content should want the best quality, as close to the source as possible, ideally the original tapes (if they have the equipment to play them), or at least a good quality transfer. I don't know if most porn studios kept tapes or reused them, because some formats would have been pretty expensive so if they wanted to save money they would have reused the tapes once the content was transferred. There are a million variables that would affect the quality of that transfer, so it may be really good or it may be so bad that a professional with any integrity wouldn't touch it. In a lot of cases really poor transfers are still being used to produce DVDs and for web content, because that is all that exists. Take one of these poor transfers, burn it to a DVD, rip it to a compressed format, upload it to a streaming site where it's encoded again, download it and pass it through AI software to turn it into "HD". You might expect the results to be quite poor, but did you think they would be this terrifying?
Try not to have nightmares!
-
@drekkin It's "alien chic" LOL
-
@NF16 yeah I see your point - and if at the end of the day the result is basically the same as some other work but just done differently, then it's a dupe.
My thinking is that so long as there are included screenshots of the material, and not just using external internet google image sourced pics, that the duplication decision can be made then by moderators. If it looks materially no different than other torrents and near same size and is just another replication with the same result then it seems to fall into the realm of what the word duplication means. That said, it'd be much more work for the moderators to validate this such potentially small idiosyncrasies the squeeze may not be worth the juice..
To complicate matters, some, albeit very few, AI enhancements are focused on other restorative functions such as vocal isolation (background music / noise isolation), grain modifications, and/or significant amount of noise alteration that the intended result is not as easily distinguishable, if at all, from still screen captures. For example, I've made a few AI enhancements to some that left visual material alone and only focused on audio alterations. On a few occasions I've made vocal track treatments because the background music was blaring or in one case (ahem, looking at you "Cock on the Rocks") that the original production had all audio recorded by a flowing pool waterfall
-
No joke, but this was shared today as Upscaled to 4K (click to view it in full 2160p).
What are people thinking when they are creating and sharing garbage like this?
