https://nyaa.si/view/1801041https://nyaa.si/view/1808448Second batch with 26-50 is out now. Full details about the upscale process are in the Nyaa posts, but basically, this is an upscale of the original SD version of SEED. It was done using a custom upscaling model trained specifically on the HD Remaster.This version is based on the SD Blu-rays from Rightstuf, so the source is much cleaner than the original DVDs, and it's from hchcsen's remuxes, so they're as good as you can get.
>>22595067SD version upscaled 200% for comparison
>>22595067Would this work with Eureka Seven? Train the AI upscaling model using the later Eureka Seven releases to upscale the original show to 4K then downscale it back to 1080p?
>>22595077Would you believe I've actually been trying this? The model "works" on E7 but the improvement is negligible imo. There are other models that definitely improve the sharpness dramatically, but they enhance and smooth the background details too unnaturally to be useable.Needless to say, training models has been a rabbit hole. I've been looking into good training references for the Dynit Sailor Moon DVDs, since my initial version with Topaz was lackluster, looking back on it.
>>22595079Is it effective on cel shows? I know there's still some old mecha that are still in SD like L-Gaim. What would happen if you upscale it with models trained from the Bluray rips of FSS or Zeta Gundam (since it's they're aesthetically similar to L-Gaim)?
>>22595095The thing is, the model has to be trained on the low res and high res versions of identical content. The problem is finding HD versions that are "good" comparisons. Unfortunately, there are many HD remasters that are overly denoised, so the model then becomes trained to remove a ton of background detail.I've tried creating synthetic downscales of 16mm content (G-Gundam), effectively making my own pretend DVD encodes, then training against the original BD version. I then also added some 0080 DVD and BD comparisons into the training to give it some real world variety, and it's pretty effective, though it can smooth some details too. The 0080 training actually goes a long way towards improving line art though. Let me grab some test screencaps...
>>22595100Okay, so here's a raw DVD screencap. I have a few models I've been working on with varying results.
>>22595111(In case it wasn't obvious, that was upscaled to 1080p for comparison)Here is my current "hybrid" model, which combines training data from the synthetic G-Gundam downscale, and 0080's DVD and BD versions. It is definitely smooth and perhaps overly sharpened to the eye, but it is the most consistent so far.
>>22595112This is with the synthetic G-Gundam training alone, which retains the grain, and is naturally somewhat softer, BUT it also sharpens the lines in very inconsistent ways at times. It also doesn't alleviate issues with the older transfer like the chromatic abberation and color bleed. On the other hand, the 0080 training data helps a lot with that, since it compared an older transfer to a newer one.
>>22595114Lastly, here was my idealistic attempt. Knowing this was a Toei production from around the same time, I figured there is no better comparison to use than the DBZ Dragon Box and the abandoned DBZ Level Sets, which were full of grain.The problem was that I was unable to find a BDMV or Remux of the Level Sets, so I had to train against a slightly denoised encode. This unfortunately tainted the training, so it removes a shit ton of detail from certain background shots, but it otherwise has a really pleasing look to the outlines (though it doesn't do much anything for the color bleed).
>>22595077If you're curious, here's the raw BDMV of E7 (US release. I'm not sure if the JP version is any better).The upscaling is aliased and edge enhanced, so it's difficult to ascertain the exact source resolution.
>>22596044And here's with the SEED upscaling model applied. I've also removed the edge enhancement and done a mild denoise before upscaling since the BD encoding isn't great.
>>22596047Well, well, well. How's this?
>>22597344This looks amazing. How does it look when it's moving? Any quick WEBMs?
>>22597363https://mega.nz/file/7F5hnQqR#t1TNVfvNkrAe_BcJol2J0u4mlIdDtSFjZH4sBNBtkxYQuick encode of the OP. The credits definitely get fucked from the upscaling, but I think it would be forgivable in the grand scheme of things.
>>22597380This is how the Blurays should've been
>>22597344Holy shit this is insane, and I thought the show couldn't look any better.would definitely pay for Blu Rays in this quality.
>>22598039>>22598077Yeah I'm honestly still blown away by how effective it can be. I've been on a fucking spree training models the last three weeks for both digital and cel purposes.
>>22598263Is it me or is it taking out a lot of the subtle shading and making it look too pixel perfect painted now? Like you oversaturated an image with a glaussian blur or something.
>>22600707I mean, it's never going to look perfect. There are quite a few models out there that make it look REALLY pixel perfect - like promo art level. Most of those tend to fuck up the blur effects used in backgrounds too.
>>22600946Oh, no, I didn't expect it to be. I'm just curious if there was anything super wonky with the upscaling (mainly with background stuff)
a heads up if people are still looking at this threadthe ever excellent ai upscale model for older web and DVD release had finally releasedhttps://github.com/Sirosky/Upscale-Hub/releases/tag/AniSD2x_AniSD_G6i1_SPAN_215K and 2x_AniSD_AC_G6i2b_SPAN_190K is what author recommend for most case,and with JaNai GUI running with tensorRT,it is ver possible to upscale video in ver high fps.
>>22602037I've been following Sirosky's guides for this whole endeavor, and these models look legit. I can't believe I'm saying that they almost seem *too* faithful just judging by the screenshots, but that's awesome that he achieved it. I'll have to try these out sometime today.I haven't played with JaNai much (I should, it would open up more models to me), as I'm always loading these in via Vapoursynth for other processing, but this might give me a reason.
>>22602298Here's the same E7 frame with AniSD. It's nice, but it doesn't feel like enough of a change to justify the compute power.
>>22602346Imo, the ideal is somewhere else. The quick model I showed in >>22597344 has artifacts in a number of places, but there's gotta be a middle ground....
>>22595067I've watched your CCA and Big O upscales and have them in my personal digital library; if you don't mind me asking, how the heck do you make these? I'm a programmer and I've been dabbling in AI image generation so I'm not a complete layman, but I'm interested if you have any resources you could share on how to get started doing this kind of stuff.
>>22606992EG is the OGHe's been doing it for 20 years.
>>22606992He did a blog back in 2014 if that helpshttps://rxmastering.wordpress.com/
>>22595067RX, question: how does your upscaling model do with text? Even my dicking around on Topaz doesn't know what to do with it.
>>22607283Same shot from the official HD version.
Well... You have already started E7 sooo......
>>22614347>>22614855Fucking spammer twat. Why do you keep doing this? Why are you so obsessed with posting that stupid rat and... whomever that traced picture is supposed to be?This is not the first time I've caught you defecating your dumb pictures on Page 10 threads.
>>22615178You do understand that picture of yours is not offensive? It's just a regular guy.There's no point, or insult, or even a zinger made there.
>>22615124>whomever that traced picture is supposed to be?How do you not know who Stephen Colbert is?>>22615198It's just silly, that's the appeal. Power to the dream as Basara sings.
>>22615210>How do you not know who Stephen Colbert is?4chan is an international website. Gringo teevee personalities are unknown elsewhere.>Power to the dream as Basara sings.Is it possible for you to reply like a normal fucking human? Or is it a compulsion of yours to add some random non-sequitur whenever you reply?
report the samefagging, spamming avatarfag every time you see it
>>22595067Moo
>>22607150Neat. Good reading material
>>22595067Hey, RX. Can you release Zeta and ZZ again without the separated files for the OP and ED? Thanks.
>thought RX retired when he dropped his blog guide >he's back with AI arsenal Confused boner.
>>22595067Would Destiny warrant the same treatment in your opinion? Or is the HD remaster of it good enough?
>>22606992Huh, I thought this thread died a while ago... I can't take much credit really. The whole infrastructure around this is much bigger and more advanced that I can comprehend, but thankfully there's a very helpful Discord community around training these models, so I followed this guide:https://github.com/Sirosky/Upscale-Hub/wiki/%F0%9F%93%88-Training-a-Model-in-NeoSRCCA is NOT an upscale. That is a legit 4K remastered source. All I did was adjust the HDR color grade. Big-O honestly warrants a do-over since it was done in Topaz, and I've actually been training cel-oriented models that could handle it much better. The only thing stopping me is the manual stitching work I did on that release to get the OP restored.
>>22606996God, watching SEED Freedom this week has really made me feel my age. I was involved in fansubbing back in 2003/2004 but didn't do EG stuff till 2006. Life is a beautiful but horrifying thing.>>22607283It's mostly okay, but text does trip it up sometimes. This particular model benefits from having a very exact HD source to train against, which isn't always an option, especially with 2000s digipaint. Topaz is impressive at times but it's too generalized to get ideal results.>>22607755Fuck I might as well just be open about it. I'm 32 episodes into the encode, remuxing OZC's subs and upscaling from the US BDs. It took me quite a while to get a model I was satisfied with, searching for sources that would not remove background detail, sharpened up well, and maintained a mostly natural appearance in the line work.I needed 2000s-era sources that had true 1080p levels of detail, clean video mastering, and equivalent low res sources as well. I settled on two movies: The Cat Returns and REDLINE. Ghibli sources are honestly some of the best, BUT, most of their digipaint works have artificial grain over them, which would taint the training results. The Cat Returns was free of grain, *and* I was able to find a DVD ISO of an older release to let the model have real sources to work with. For Redline, I only found a BD source, so I ended up creating my own "DVD-equivalent" MPEG2 encode that was used as a synthetic source to train against.>>22621058I hate Destiny too much to bother.
>>22621509Some pics of the upscale
>>22621583
>>22621584
>>22621585
>>22621588
>>22621583>>22621584>>22621585>>22621588>>22621590Looks good. Will probably replace the OZC encode that I use.
WHO KEEPS LETTING RX ESCAPE THE RETIREMENT HOME
>>22595067Thanks RX. I hope you never fully retire and keep blessing us with your releases even if they're just experiments.
>>22621509>I hate Destiny too much to bother.So say someone else wanted to give a go at Destiny in your stead. Do you think using one of those pre-made Sirosky models is probably good enough or is it better to go through the motions of training it against Destiny HD like you did with SEED HD? If a custom training job is better, you mentioned you trained your model on a handful of episodes, but was there anything about those that made you pick them or just that they were spread across the show's run?
>>22622255Different OPs/EDs, different landscapes (space, desert, cities, etc), and just more variety in general. The more examples you can give the trainer the better.
>>22622266Gotcha, that makes sense. I have no idea if I'll even be able to make any of this stuff work, but after getting halfway through your SEED upscale, I don't see myself ever going back to the official HD version if I don't have to.
>>22622328I mean, you could try the model I used for SEED on Destiny if you want. I'll share it when I get home later.
>>22622336That would be awesome. Certainly can't hurt to try.
>>22622349Make sure to grab the SDBD remuxes if you can. I think hchcsen has them on Nyaa. Also, Destiny *should* mostly be 23.976fps, so you likely can get away with a simple inverse telecine on it.
>>22622356Yep, got those already so I should be set there.
>>22622369https://mega.nz/file/iJYQzKJQ#Jc1hIeO1ESwhsESp4rwZjCM4yFmMG9uo2XHK1xZmicEHere you go. Godspeed.
>>22595067This is more useless than the HD remaster
>>22622894Appreciate it! Here's hoping it bears fruit.
Oh EG, you're stilling doing anime releases? I thouht you retired from this for an encoding job with a company or something.
>>22623039I left that company in 2015. Been an editor among other things since then.
This has probably been asked before but what's your rig?
>>22623053It's getting long in the tooth, but I have other things to worry about before upgrading (unless a gig necessitates it of course):- Ryzen 3900x- 32GB RAM- RTX 2080 Super- Win 10
>>22623047Oh. Time marches on, I guess. Was that encoding job any good? Were you being asked to coform to shitty industry encoding practices?
>>22622894Well, gave it a shot with Episode 1. Seems like it works pretty well to me. It feels like Destiny is a little noisier at the source compared to what your SEED comparisons looked like, but I guess that means it's accurate.Comparison of 3 random shots - https://imgsli.com/MjYyNzMx
>>22621509>God, watching SEED Freedom this week has really made me feel my age. I was involved in fansubbing back in 2003/2004 but didn't do EG stuff till 2006. Life is a beautiful but horrifying thing.Bro I know people that have archived all your anime releases. All of them. You have left your mark on fansub history.
>>22595067can we get an AI dub for freedom using original voices.
>>22595118A bit out of topic but I wanna ask, any chance of you experimenting for removing the pink tint on Sailor Moon?
>>22623248You can always give it a very gentle denoise if you feel it would benefit, but I think it looks pretty natural. This model wasn't really trained to denoise at all since the SD source was also very clean, but other models offer more of a denoising and general repair approach. If you ever want to try training, you can actually build off of existing models, so you can get a head start by using my SEED model and adding Destiny training data to it.>>22623297I appreciate that, but I barely worked in true fansubbing. EG has always been a rip group. My brief fansubbing stint was with a group called Anime-Blitz for about two years. >>22623302You'd need a completely different approach to do this, but also raw source audio of the original VAs.>>22623307I did actually try, and I even put out up to Sailor Moon S, with varying results. I created a LUT for every season that took different approached, but I was never really satisfied with the result. On a related note though, I recently trained a model that gives a pretty impressive upscaling result to the Dynit DVDs.Sample Encode:https://mega.nz/file/3J4V2QpD#DE2OJ1B9354sFfGjzPbgSdOem8usNKv24dW_tL4f9nA
>>22623424DVD Source.
No offense to your hard work, but wouldn't you be better of training AI models on multiple film grain samples? Then use AI to remove film grain from the releases.I own several animation cels, and no grain is present. The original animators did not intend for their work to be covered in grain. It's merely a consequence of using old school film.
>>22623473Putting aside the extra computing power to do this in two stages (you'd ideally just get a model that can do both), it would be extremely difficult to get perfectly clean references. There's a difference between training a model on something that's denoised and something that's inherently clean. What you'd basically need are the actual cel references (complete with backgrounds).I've tried training against denoised sources, or sources that were HD but clearly DNR'r, and all it has led to is training the models to aggressively denoised the SD image. Sometimes it's tolerable, but I still don't find it natural.When you start going along the the thought process of "it's not what the animators intended", I absolutely understand where you're coming from, but then you have to start questioning every other issue with the production techniques of the day. Should the image also be 100% stabilized and free of gate weave? Should mis-painted cels be corrected? Should the brush strokes of certain cels be smoothed out? I doubt they intended those issues to crop up too, but it's how it was produced.The answer isn't black and white of course. There are different approaches to remastering all the time, and in the world of actual professional high-end film restoration, the results can be sharper and clearer than was even possible back when it was produced. And yes, I agree that we should take it to that level *if it looks good*... But it often fails to.
>>22623505Different anon here. I'm curious. They have video packs of recorded pure film grain. 8mm,16mm, 35mm, etc. Plus different brands like Kodak, Fujifilm, etc. Couldn't you train your AI with those packs to recognize those film grains? Then ask the AI to remove the grains from footage?
>>22624193It's not that simple. Grain packs are also neat but grain itself doesn't apply universally across all scenes. Bright and dark shots render differently on film stocks, and then you have to consider that most BDs have at least some form of DNR. There are a few that are seemingly untouched (Outlaw star, Escaflowne, and NGE come to mind), but it makes it very difficult to create a model comprehensive enough to detect what kind of grain it is and remove only that. Could it be done? Perhaps, but it's beyond my abilities and resources.
EG, for the Big O release and that stuff about the OP being interlaced 60 fps, why couldn't you just use ordered chapters for the OP?
>>22624317Hardware compatibility, too many people bitching about ordered chapters through the years
>>22624263Hold on. Whats the issue? You can train your AI using grain packs recorded in both light and darkness. There are packs of those. Different grain stocks aren't an issue since these packs contain so many different stocks and brands.>DNR If something has been DNR'd pre-release, then there's nothing you can do about it correct? Not unless you can find the original pre-DNR footage which are stored in vaults and archives. So your baseline is basically the DVD and BD releases. And these days sometimes only the BD release is the only one. Don't get me wrong. I appreciate the work the work you you've done. You are pretty much a legend in the fansub and release community. I'm just confused about your goals on this one specific issue.
>>22625923Alright, let's assume you could train on every possible baseline expectation of grain. Now the model is essentially trained to look for "perfect" grain as defined by the grain packs. But if something is DNR'd, its grain is no longer "perfect". So it's not like it can then do a simple subtraction of grain from the image. Combine this with the fact that films can have duplicate grain structures (35 to 16, and even 16 to 16mm), and there are countless reasons it's not that easy.
>>22625923>>22626003Also to expand on this further, I don't know how to train on only one reference. The way these trainers work is you feed them the "problem" and the "solution". So for example, you can show it content that is aliased and the matching references that are unaliased, training it to fix aliasing.For grain, you'd have to show it what the "proper" result should be. For a grain pack, would it simply be a solid grey background? It has no idea how image detail is comprised beneath the grain.
>>22626003Wait. Hold on. Aren't you training the AI to recognize individual grain particles? How would grain video packs be insufficient? Even if a release has ubdergone DNR, my understanding is that the DNR process removes grain particles based on a mathematical algorithm. And that you can adjust the strength of DNR depending on the software. But DNR doesn't physically change the "size" of grain particles already present in the footage. It just removes them.But if you use AI, you can feed it multiple examples of different 16mm film grain packs (dark/light, night/day, super 16mm/regular 16mm, Kodak/Fujifilm, etc) under different conditions. Eventually the AI learns what 16mm film grain particles generally looks like.Then you can ask it to remove film grain particles from an image. Of course Im summarizing a lot here, but that's my understanding.
>>22626041That's not true at all, DNR absolutely can alter the size. On a basic level you are averaging neighboring pixels, so your averaged grain will in effect be larger.
>>22626003>>22626165Different anon here. The grain packs aren't just one type or size of grain. An 8mm grain pack will have grain of various sizes typical to 8mm film. The 8mm grain isn't just one size. It will vary. Same with 16mm and 35mm. If dnr reduces size in addition to removal, then that reduced size grain particle should still be recognized correct? Only the AI would think a 16mm grain particle that has undergone dnr might be a 35mm grain particle.
rewatching it now and honestly I kinda prefer the OG frames compared to HD remaster. The only "oh fuck what were they thinking" was athruns. Kira looks a lot better in the original frames compared to the remastered "destiny-like look"
>>22626236like i think the biggest problem is Kira (Seed) and Kira (Destiny) have a distinct difference with their hair and look. (Seed is more baby face, hair is more flatter) but the redrawn scenes use the Kira from Destiny as a reference and it throws you off.
>>22623424>>22623426Is Sailor moon ever going to be released in true HD? The show was made in 16mm film right? That's good enough for 4K. Toei made it right? Why is it taking so long. Sailor moon is iconic
>>22628685Toei seems to not give a shit about their greatest legacies
>>22628685They did release BluRays not too long ago, but they look as bad as you'd expect from Toei.Despite how big Sailor Moon was in the 90s, it seems like they do everything in their power to not give a shit about it. Just look at Crystal. It's a dead property at this point.
>>22629142Holy shit lol
>>22629142>>22629230From the Dynit upscale
>>22629242Yeah, it's insane how the upscales you did before are somehow much better.It's honestly a miracle that those Italian DVDs were ever a thing. Can only hope that the same distributors make a remaster of their own, but I seriously doubt that will ever happen.
>>22629142That can't be the original film scan. It's blurry as heck. Its a DVD upscale
>>22629411Thatsthejoke.jpg
Are there any additional scenes added in the movies/special editions, like the Mu's Perfect Strike and the Experimental Guaiz, that you would consider merging into a release?I've always hated having nice extra stuff locked behind movies. Having a definitive everything included series would be cool. I understand audio and sub timing would be a little annoying, though.
>>22629943
Did hchcsen delete fucking everything? All there torrents are gone.I wanted to compare this to the SDBD remux and it looks like everything they did has been scrubbed from the internet.
>>22629423Not that anon but I don't get it. I'm not involved with fansub drama. Please explain.
>>22630892No sub drama. It's about the actual, official release.
>>22630907Again I don't get it
complete newfag here starting to get interested in gundam for the first time - where's the best place to start?
>>22630909Too bad.
>>22630942I literally politely asked for an explaination. If you don't have one, then why are you even posting? I would like to speak to someone who is actually knowledgeable about all these insider memes.
>>22630911The start.
>>22630968See >>22629423
>>22630968Yeah I don't get it either. I guess Toei did a bad scan of their old film. That's it
>>22630968>>22630993They tried to eliminate grain and just fucked up the image.
>>22630887Sure looks like it. There's one that's still tagged hchcsen that was an anonymous upload and probably couldn't be removed but yeah, it's all gone.
>>22630999I don’t think people should try to eliminate the grain.
Is it just me or is the dub track not included in this release?
>>22630976it was good I liked it
