Introducing the Society for the Preservation and Restoration of Gokkun Classics(?)
Hi! I like porn, especially Japanese gokkun videos that haven't been new for 15-25 years. I also have two 4K monitors, and my favorite smut looks like butt on them. If you're anything like me, then read on, my new friend!
Around a decade ago, I used to rent DVDs to make my own rips, and when the titles I wanted weren't available for rental, I would buy and resell them on the secondhand market. In short, I wanted to enjoy my favorite titles in near-DVD quality, but I didn't want to deal with a physical porn collection, let alone spend a small fortune on one. At the time, I used legacy Handbrake settings I thought were an ideal compromise between video quality and file size, but my laptop's screen was hardly high resolution, and I'm pretty sure my external hard drive was no larger than 512 GB. Even when I got around to buying a Full HD TV, I still thought DVDs looked pretty okay on it.
View attachment 3438774
A DVD rip compressed with AVC (H.264), then later compressed again with HEVC (H.265)...because hard drive space was at a premium. (Note: all images are compressed JPEGs optimized for the web.)
Fast forward to last spring, and I finally bought my first 4K monitor and realized just how horrid DVDs can look on an Ultra HD screen. I had seen AI resolution upscaling and frame interpolation do some pretty impressive things, so I purchased Topaz Video AI in the hopes of breathing new life into my aging digital media collection. Here's a peek at what I've been up to.
Although I used Handbrake's default deinterlacing settings for any DVDs that I ripped, AI resolution upscaling can really hang a lantern on any lingering combing artifacts that you otherwise never would have noticed. So, the first step I take is to throw TVAI's "interlaced progressive" models at the compressed DVD rip. (Assuming I don't need to edit things in Final Cut Pro first, that is, such as combining multiple discs into one video.) I leverage every appropriate model that makes an appreciable, non-redundant improvement, then blend the output evenly in FCP to balance the strengths and weaknesses of each model. This typically means creating a three-way blend of the Dione Robust, Iris (Medium Quality), and Proteus models. Since the Proteus model has become available as an "interlaced progressive" model, blending at this stage probably isn't so necessary anymore, but when working from such low quality sources, I'll take even a 1% improvement.
I should mention that when using FCP for the first time, I take the opportunity to raise or lower the volume as needed in 1-db increments (for simplicity's sake) so the audio is about as loud as it can be without distorting. You might imagine this typically means having to raise the volume, but more often than not, I end up having to lower it. I used to let FCP auto-correct issues it detected in the audio by boosting loudness, filtering noise, etc., but I eventually realized this caused more harm than good, especially to music and any non-speech sounds. It's a shame I can't do more to doctor audio, as the quality in porn tends to be quite poor.
Also, I haven't been able to find a way to automate color correction that produces consistently good results, so I don't typically muck around with the color. But I've been experimenting with what I can accomplish in Pixelmator Pro on this video and I think I've found a preset I don't totally hate. I tend to prefer warmer, more saturated colors, and when watching the color corrected video side-by-side with the original, I can't deny that I prefer the color corrected one, especially on my OLED. If this step becomes a permanent fixture of my workflow, I'll have to perform it after blending the deinterlaced models in FCP when files sizes are still relatively small, as PP struggles with large videos. At any rate, just prepping videos takes the better part of a day on my rig.
View attachment 3438775
Left to right: the twice compressed DVD rip vs. an even blend of three "interlaced progressive" models after color correction. (Note: click to enlarge.)
Once I have a cleaned up video in the original resolution, I go back into TVAI and perform a 2x upscale on it, usually using the Artemis (Medium Quality), Iris (Medium Quality), and Proteus models. Artemis produces the sharpest resolution upscale, but also leaves the strongest "AI upscale" impression. Iris, meanwhile, produces a softer resolution upscale, but also produces the finest detail in hair and maybe skin. Proteus tends to be a good balance between the two, so it provides a good base to improve on when blending these models in FCP. This step, too, takes at least a whole day.
Unfortunately, it's pointless trying to go beyond a 2x resolution upscale, at least when starting with a compressed DVD rip. Believe me, I've tried, but no amount of pixel peeping could reveal any discernible difference between a 2x and a 4x upscale. Also, resolution upscaling seems to blur rather than sharpen censorship mosaics, but blending can help alleviate this.
View attachment 3438776
Top to bottom: 2x upscales using the Artemis (Medium Quality), Iris (Medium Quality), and Proteus models vs. an even three-way blend of each—the differences are subtle enough they can be difficult to discern in side-by-side comparisons, but compositing readily reveals them. (Note: click to enlarge.)
Once I have the best 2x resolution upscale I can produce by blending different models in FCP, I go back into TVAI once again to double the frame rate. Chronos has been my go-to frame interpolation model, but I've had so much trouble with it over the months, I think it's time to switch to Apollo from now on. As finicky as these frame interpolation models can prove, I personally find high frame rates more impressive than high resolutions, so this is where the real magic happens for me. Still, this can take a day or two, potentially more depending on the video's length.
View attachment 3438777
Top to bottom: a series of four frames using the Apollo model—without checking the source, I wouldn't be able to tell which two were original frames and which two were interpolated frames. (Note: click to enlarge.)
I'm something of a realist when it comes to these things, so my personal goal is to bump the quality one rung up the ladder. In other words, make VHS look like DVD, DVD look like HD, HD look like Full HD, and so on, albeit it with a buttery smooth frame rate. After all, an AI remaster of a DVD can really only pass for native HD at a glance in close and maybe medium shots, and even that's assuming a decent source was used. Wide shots are pretty well a lost cause, as there's just not enough detail in the original image for these models to work with. So, as a last step, I create a 720p 59.94 fps mp4 in Handbrake using HEVC (H.265) compression, just as I would if I were ripping a Blu-ray disc. This also affords me the chance to crop any junk from the overscan area as needed in 8-pixel increments (again, for simplicity's sake) without worrying about the target resolution.
View attachment 3438778
Top to bottom: the twice compressed DVD rip (enlarged) vs. the final AI remastered video, an imperfect imitation of native HD but a dramatic improvement over the source.
Still, I'm beginning to realize that if I'm going to go through all of this effort, spend all of this money on software, and cause this much wear and tear to my setup, I may as well do it right. And that of course would mean using the highest quality video available from the start by going back to the physical DVD discs...which is where you might come in. If you live outside Japan, getting your hands on DVDs to create uncompressed ISO images may not be nearly as difficult or as expensive as you might imagine. And who knows? Maybe even 1080p will become attainable.
So, I'm posting to gauge interest in forming a group to preserve and restore these DVDs while maximizing mutual benefit and minimizing individual cost. Preservation means saving what few physical DVD discs remain from incinerators, landfills, and disc rot. Restoration means everything I went over above. I'd like to concentrate our efforts on the stone-cold classics from the naughts before attempting more vintage or more modern titles, as that's where the greatest good can be done. If you're at all interested, send me a DM for details. I'll check my messages at least once a week over the next four weeks or so before peacing out. But if there's enough interest in pooling our collective resources to justify the effort that would entail, I'll make the necessary arrangements. Otherwise, I'm happy to keep what little I can accomplish on my own to myself.
Just bear in mind this would not mean free porn, as contribution would be the price of admission. And besides, this stuff is already pirated and available in the wild, including right here. Nor, for that matter, will I be soliciting monetary donations or selling a service. Rather, think of it as loaning a long out-of-print book to a private library, and by doing so, gaining access to all the other long out-of-print books that everybody else has at one point loaned. Don't think about this analogy too hard. It's a magical occult library that doesn't require books to be donated or returned after being checked out. Moreover, that library takes that long out-of-print book and lovingly digitizes it into a high quality e-book orders of magnitude superior to all those old, crappily scanned PDFs widely circulating elsewhere. And if watching that library slowly but surely grow while breathing new life into all of our old favorites inspires you to track down even more long out-of-print books to loan and in turn enjoy on your favorite modern devices, then you're starting to get the right picture.
Cheers!