30

Mahou Shoujo Ikusei Keikaku (BD) – Volume 1

Happy new year, future-people.

Apparently MahoIku comes in 4 volumes of 3 episodes. Interesting. Also, yell at eXmendiC if filesizes offend you. :^)

A further note here: Certain in-progress projects will not continue (Charlotte, Haifuri, Tanaka-kun, and so on) until my hard drive data is returned some time in January. I can, however, start new projects as I am not reliant on project notes left on the drive.

Volume 1 contains Ep01, Ep02 and Ep03.

Mahou Shoujo Ikusei Keikaku (BD) – Volume 1 (BD 720p): [Torrent]

If you encounter any playback issues with a GJM release, please make sure you are using a recently-updated version of CCCP or KCP.

Good Job! Media

30 Comments

  1. “Certain in-progress projects will not continue (Charlotte, Haifuri, Tanaka-kun, and so on) until my hard drive data is returned some time in January.”

    Charlotte will probably not continue either way… right?

  2. Well, you’re not dropped them, are you ?
    Thanks for Mahou, I can begins the show with BD 🙂

      • Every time I hear that comment it makes me wonder what exactly they mean by it. Sometimes its due to purity of the work. (Which BD’s are imo mean’t to give you better experience) Right?

        Though I’m still curious as to what it looks like.

        • Source resolution mastering has a part in it. Some shows don’t lend themselves well to upscaling through any method. They can come out aliased, blurred, with loss of details, and so on.

          Some encoders don’t really bother about this and do it anyways, but others take a more pure approach to their work and prefer not to provide something they know will look worse than keeping the work at its studio-mastered resolution.

        • http://diff.pics/xxV7SdcLGDFC/1

          1st screenshot: The source has a “special upscale”. I call it OS Bicubic (or oversharped Bicubic). I proper downscaled it back to the native resolution (720p).
          Also there is heavy Aliasing, which I had to fix with eedi3 (so the ED has some detail loss).

          2nd screenshot: The source has a normal upscale this time with a bit Aliasing. So I proper downscaled it to the native resolution as well and fixed the Aliasing after that. You can see that my 720p is sharper with a decent resizer (nnedi3) than the source.

          3rd screenshot: The same as the 1st screenshot, just without the Aliasing part

          As you can see, there are no benefits of a 1080p version EXCEPT I would upscale it back (like on the screens) with nnedi3 to 1080p. That means: 1080p -> 720p (Debicubic, Debilinear) -> 1080p (nnedi3).

          Does downloading a 1080p still make sense that way?
          If you have a toaster, it does. But if your PC is good enough to handle nnedi3 or a similiar upscaler, no.
          Some groups don’t bother with that and release a 1080p as it is (no downscaling to 720p and then upscaling to 1080p). That literally means a blurrier video without benefits. Of course doing something like that is only useful if the native resolution is under 1080p (like for MahoIku). But I also know some groups that really care about the 1080p and do it that way (DameDesuYo for example).

          Just make sure the 720p is Hi444PP (4:4:4) and the show isn’t above 720p, so you really have no quality loss against a 1080p. And it shouldn’t be downscaled with Spline36 or something similiar to 720p. But without being the encoder, you can’t really check that except comparing it with the source.

          I hope that explained it a bit.

          • I believe, I understand now. Though I know this will probably annoy you or anyone but I’m still very Curious as to how it would look like in RT.

            I do see where your coming from. Thanks @eXmendiC

          • Can’t nnedi3 only upscale by factors of 2? So are you upscaling to 1440p and downscaling back to 1080p with some other scaling algorithm?

            Also, I’ve always wondered how would an encoder determine the b and c parameters that studios use to upscale to 1080p with a bicubic algorithm? Is there a way to actually determine them for a true lossless downscale to 720p with debicubic? I’m just thinking that if the downscale isn’t truly lossless then what’s the point of downscaling then re-upscaling to 1080p when you can fix most of the issues without downscaling? I guess how the final video at 1080p should look is subjective to the encoder, but if the downscale isn’t lossless, then re-upscaling to 1080p with a different algorithm isn’t how it _should_ look, but rather what the encoder thinks it should look like.

            I do think there are still some things that get blurred after downscaling like credits, grain/noise, and other things as seen in your screenshots (although some may be just from heavy aa). Some lines and stuff can also potentially look worse (darker, thicker, ringing, etc.) than just keeping it at 1080p and applying mild-to-moderate aa.

            Although, I guess I agree for shitty BDs maybe it’s not worth the 0.1% difference and larger file size to encode at 1080p, except for the people who want closer transparency to the BD (with issues fixed of course). I guess my point is that there can be benefits to encoding in 1080p, but they’re minuscule and perhaps not many people will notice.

          • “So are you upscaling to 1440p and downscaling back to 1080p with some other scaling algorithm?”
            nnedi3_resize16(1920,1080,kernel_u=”Spline16″,kernel_d=”Spline16″,invks_u=false,invks_d=true,invkstaps=6)
            For madVR just use Jinc, SSIM or Bicubic150. The results are probably pretty similiar, maybe even better.

            “Is there a way to actually determine them for a true lossless downscale to 720p with debicubic?”
            Well, you can’t really say which values are right. But most of the time c=0 and b=1 works fine for it and fixes everything, while other values don’t.

            “I’m just thinking that if the downscale isn’t truly lossless then what’s the point of downscaling then re-upscaling to 1080p when you can fix most of the issues without downscaling?”
            If you talk about releasing it in 1080p, then guess why I haven’t encoded a 1080p version. It’s only good for people with a bad PC that can’t even handle Jinc or nnedi3 as an upscaler in madVR.

            “I do think there are still some things that get blurred after downscaling like credits, grain/noise, and other things as seen in your screenshots (although some may be just from heavy aa)”
            Credits are usually in 1080p, so you can never get them lossless in a 720p version. But I don’t really care about credits. I removed all the grain/noise of the source and added own static grain (some people may dislike that, but I like to watch my anime like that). But they are usually 1080p as well. And yes, it’s because of the heavy AA. Eedi3 is probably one of the strongest AA filters, but I had to use it for the ED. No chance getting around it. The other part of the show was simple nnedi3 AA.

            “than just keeping it at 1080p and applying mild-to-moderate aa.”
            That isn’t a good option if you have an OS Bicubic source, really. You can maybe do that if your source is Bilinear/Bicubic with some Aliasing, but I still rather downscale it to 720p (if the source is 720p).

        • While playback there shouldn’t be any difference between the 720p and the 1080p (upscaled from 720p while filtering). You can decide the resizer in madVR. It’s hard to take pictures like that, because I actually had to encode a 1080p version then.
          But I have a “Source vs. Encode” comparison here: http://diff.pics/1iLWiTQGAKjl/1
          I’ve used nnedi3 for the 720p to 1080p upscale, which is also available in madVR.
          Note: I haven’t used a sharpening filter, that’s nnedi3.

          The other comparison was a “Source vs. Preview of the filtered Source”.

          • At the end of the day it all comes down to tastes and the encoder would prefer it to look.

            I pretty much discovered this myself.

  3. At the end of the day it all boils down to what the encoder would like for their encode.

    I pretty much discovered this myself and will experiment like I usually do. Especially since recently I just got myself a quadro. :/

Leave a Reply