/a/ - Animu & Mango

>>907710

Whatever its called.

As long as its the full capture.

I hope you faggots are using alpha channel blending to ensure you don't leave ugly seams and jagged sparkle/water effects everywhere.

Fixed the Clock

Is there a better way of composing these. My normal workflow:

- Screenshot a series of frames in mpv

- Lower-most layer entirely black

- Align them as best as they can be, there's usually subtle distortions

- Set all layer blending to "lighten only"

- Layer mask to remove any obvious seams between frame boundaries, which can happen if there's movement.

The "lighten only" usually fixes colour differences between frames which can create very noticeable banding otherwise.

>>907799

If there's a better workflow, I'm not sure of what non-trivial improvements there are. What I do differently to you is skipping the lighten-only blending and the bottom black layer, and having layer masks do most of the work. I try to put a fair bit of overlap between each pair of snapshots (25--50%) so that the layer masks can have slowly-ramping gradients. Sometimes a frame is shifted a fractional number of pixels away from another frame and a slow ramp deals with that nicely. It also solves the banding problem in most cases, but I've run into images that needed manual adjusting with a few nearly-transparent, soft-brush addition or subtraction blending strokes. Pic #1 related, it was a hard one and I fucked a few things up in the background. Of course, if there is movement between frames, I manually adjust the layer masks afterward.

Diagonal pans are a much more difficult case (especially if you want as much of the edges filled as possible), one which I haven't quite figured out a consistent method for. The usual method of aligning each frame to the previous one has the flaw of minor displacements adding up to multiple pixels by the time you get to the end of a series of frames. This has to be fixed manually by hiding every frame except the first and the nth one, and aligning the nth one to the first, for many values of n. This is a long and tedious step. Technically, the usual horizontal or vertical pans have this problem too, but it doesn't matter much when there are only 5 or so frames. The advantage of the usual method is that a layer mask can be drawn for the second frame extending the first frame, and then the mask can be copy-pasted for the third frame extending the second, etc. (but make the mask cover more than necessary in case a later frame is x+1 pixels away from the previous frame instead of x pixels)

The second method is where I align the last frame to the first frame, then put the 1/2-way frame underneath, then the 1/4 and 3/4-way frames, stopping when I've gotten as far as I want to. This has the advantage of getting the placement right from the start as well as being friendly to cutting it off early. It's a lot harder to manage the layer order though, and takes a lot more manual working of the layer masks, so much so that it's not really worth the advantages. Laying out both methods like this makes the first one sound better, but I think I need to experiment more to tell. I don't remember which one I used for pic #2.

When a pan also incorporates zooming (Pic #3 related), that just makes things need that much more manual work. Use the most zoomed-out one as the base and downscale the other frames as necessary. You basically have to give up on not having fractional pixel displacements between each frame, so just manage as best as you can with slow ramps in the layer masks. It seems to me that most zooms are linear, so you can safely linearly interpolate the downscaling for intermediate frames if you have an initial frame and a final frame. Sometimes the initial frame and final frame have no overlap, so treat it as two series of frames that do have overlap between initial and final frame. The workflow is similar to the second method for diagonal pans, but with scaling the frames instead of displacing the frames. Do it that way instead of in frame order because you absolutely do not want minor errors in zoom to multiply together, that will look a lot worse than minor displacement errors adding together.

(only last is NSFW)

>>907845

Great post anon.

>lighten-only blending

This is just the straight-forward way. Found that trick when doing #1 since even with lots of overlapping frames there was serious banding. As for why the frames had varying colour, maybe because the source was a TV broadcast. It may be less bad with BDs.

>minor displacement errors

I've found with most images that it's more that one displacement cancels another out, and the image as a whole fits well together with only some erasing to get rid of bad seams. Mind I've only ever done the simplest stitches. But that Kobato one was annoying for having variations even with the same animation frame, and having to sync up to the same one because she's breathing lightly.

>Diagonal pans

I've never tried one of these but I'd probably not bother going totally crazy at keeping the edge details, because it's the main girl(s) that are important. It also means there's more to fit together.

>zooming

I'm totally wanting to see if I can do one that includes this AND rotation now you bring it up. Trying things is educational.

>>907925

>karen

More where that came from.

>>908037

Thanks a lot.

>>908039

Flat is justice.

>>907707

Why girls? Girls are for gays.

Challenge accepted on making a harder splice. No way I would've managed this in GIMP, but thought up a particularly ridiculous way: Georeferencing! The hard bit was actually getting the transformed images back out so I could do the usual splice routine in GIMP.

>>908182

I prefer big, which is why she's best Nep because she caters to both.

Nepgear's turn, and video of workflow for this too since I think this deserves documenting, and I don't think this sort of image transform feature exists even in Photoshop. Two details I forgot to mention: Change filter to "Cubic Spline" or generally "what looks best" because Lanczos produces jagged linework sometimes. Also for point picking try spreading them across the image if possible, and while 4 is the bare minimum you should try for 5 or more to detect gross errors or if two frames have significant differences.

Also there's different transforms, and some will actually not do the image uniformly but you need a lot more distinct common points between the two which most anime don't have, and there is no validation against gross errors except "this image is distorted hilariously".

Better version of the Ploot too. The test example was done from start to finish instead of the reverse and her head ends up noticeably distorted. Plus Lanczos results in terrible lines.

>>908456

>Lanczos produces jagged linework sometimes.

It's a better resampler than keys cubes. The jagged linework is likely ringing because for some reason developers love to configure it to be overkill sharp by default. If its parameters can be changed I recommend doing that and sticking to lanczos.

But your editing is very good anon, and I've never heard of that editor until now.

I'm not very good at this stuff, in all of them I only made sure the images lined up on GIMP, and nothing else.

At the very least I made sure the jpeg encodes were good.

I have more than I thought.

And then ImoImo.

I'm now regretting not having a folder for these, since I have a few but they're scattered all over my HDD.

>>910063

That's because it's a program intended for GIS work, and that plug-in specifically is intended to fit things like aerial photos and maps to co-ordinated points in a database. It just happens to also be useful for making lewd Nep pictures. I'll do Vert in a bit but for now, not one I made.

Unfortunately you cannot configure the resamplers, you can just choose one, similar to GIMP. Also I'll point out that QGIS is free and open source so anyone can make use of it, but I'd only advise it for hard stitch work.

>>910067

Angels!

I hope you like them.

I swear there was a Vert pan that required stitching but her scene zooms out. Did these instead. #2 was annoying for the fading glow necessitating a LOT of blended frames. Also stopped doing the "lighten only" thing it's sometimes bad. Instead I just do a gradient along the edge of the layer mask and copy-paste it to all frames.

>>910224

Yes.

No idea how I doubled that up.

Last one doesn't look that great, but i'm posting it anyway.

>>910284

The easy way is basically cross-fading assuming you have enough frames to work with. Layer masks let you non-destructively remove parts of a layer so you just apply a black-to-white gradient on edges to start out (black obliterates its corresponding pixel, white keeps it, grays make it translucent). However it's way more powerful than that so demo video I'm cooking will also have to fix parallax since it's not just colour seams but parts plain not matching.

Video done. 20/20 hindsight says I should have just cloned stuff from another layer with the one cord spanning most of the image, Too late for that though and the lazy solution would've been just cropping the image and ignoring it. Hope it's helpful.

>>910284

Why do the frames vary so wildly in colouration? Regardless my attempt. Purely lining them up, gradient masking, and setting the layer blend to "lighten only" since it was much easier.

>>910321

It's perfect, thank you very much i appreciate it. I guess the lighting changes by frame because it's a transition scene in the anime. Again thanks for taking the time to show me the ropes, i really appreciate it.

>>910362

I hope you were at least trying to spread it as a concept because unless you like tinkering a lot you don't normally figure things out on your own, and I've learned a lot from generous anons with things like WebM creation. It was at least a couple years before I picked up on layer masking, prior to that I was entirely relying on "lighten only" and direct editing/erasure. The former was bad for this particular attempt since the water isn't static and it ends up overly bright.

>>910327

A word about "lighten only" is that it should never be used if the linework does not almost perfectly overlap. This was the case for the Nepgear one further up for example and you see parts of it get obliterated with that. It also requires the lowest layer be black and covering the entire scene.

Just because, the other three. Purple was trivial. Black gave me the most grief with certain cord joins so it's probably a bit iffy. White, so-so.

>>910362

Unfortunately what's also mainstream now is a different speed of movement for the back and foreground.

>>910737

That's what clone tool or re-drawing is for. You can use "sample merged" to make it take the image into account, not whatever layer you set the tool onto. Sometimes though selective masking between frames does the job as in pic related. I was actually trying for an animated stitch with this but it might be far too much work because the background shifts between assembled frames ridiculously

>>910816

>clone tool

>redrawing

That's far too much work.

>>910869

Also even with actual re-drawing it can be fairly easy. This isn't even a great one but it's passable.

https://my.mixtape.moe/xtachb.xcf

Is there a way to stitch Marie?

It may be difficult because of the background but its worth a shot.

>>911922

Had to be rescaled and didn't exactly match up, but it wasn't too hard to connect them around the neck.

>>907799

Pretty autistic, i just printscreen and crop with paint

>>912853

You're just shooting yourself in the foot. MS Paint having no layer support is reason enough to use GIMP. But even for simple pans MS Paint is often inadequate or more work. Most basic stitches have colour seams between frames which are trivial to fix. But then you get to things like parallax, frames that need transformation because of scale or rotation differences, and worst, re-drawing. For example >>911829 has her hand move off her leg as it's panning, her right knee is re-drawn among other corrections. This also briefly has her not holding her whip sword up, but it happens to be easy since the whole thing is visible in that form. No re-draw. Just seam correction and minor parallax seams on Rei.

Also do you actually mean a stitch, that is a composition of multiple screenshots to one seamless image?

>>915685

Lots of subtle details I like.

Some Bestia. Request taking still open.

>>915871

That's supposed to be a loli?

Disregard #2 it's got some errors.

>>916082

Yes I did, and the randoseru does not match her shapely body. I don't get it.

>>916557

Lead us not into temptation for bad anime. Too late though.

Rare stitch.

>>916602

Yes, i just finished watching Hajimete no Gal. It was pretty shit, but whatever i got some nice screenshots.

>>916630

That you did. There's something to be said for teasing content like that.

>>918556

>#4

>NSFW

>>916082

Just because someone has a red backpack doesn't make them a loli.