frio

Alternatively, control-B to turn the "zoom image to window" off right after making a new image. Zooming manually also does that.

Same deal there, the files are just completely blank, a couple megabytes of zeroes. Whatever file restoration happened there was not very successful.

That example file is all blank, full of zero bytes. There's not even a trace of anything to rescue, I'm afraid.

Paste Warp+ reshapes whatever's on the clipboard into the current selection. Original on left, random selections paste-warped into on the right.

The gradient tool uses dithering to reduce the amount of color banding. The closer the colors are to each other, the more intense the dithering becomes and alpha counts as well, so alpha 255 to alpha 0 are maximally far from each other, but alpha 70 and 0 are much closer and get dithered harder. When you change layer opacity, there is no dithering, and you get the banding instead. Here's a synthetic comparison: top part is a gradient from RGBA 255, 255, 255, 31 to 0, 0, 0, 0, on a black background. The bottom part is a RGBA 255, 255, 255, 255 to 0, 0, 0, 0, with layer opacity set to 31. After merging the layers, the visibility was boosted with the levels tool to bring the 31, 31, 31 to 255, 255, 255. As you can see, the banding is intense on the bottom one, whereas the top one is very fuzzy but you can't really tell the banding. It's a tradeoff and you kind of just have to choose which is preferable for your use.

The ZIP filetype plugin lets you save all layers as a zip of pngs.

BoltBait's Fill -> From Clipboard does exactly this and respects the alpha channel on what you're filling with.

New version crashes paint.net with some plugins, notably Nvidia normal map plugin. It's very old and Nvidia no longer has a download for it, but this is probably a legitimate enough source (the installer checks clean on virustotal.com): https://gamebanana.com/tools/347 There were also others that crash with the same error but are commercial. Others were okay. Crash log:

There's a partially transparent gradient at the bottom right, it's what is being detected as an edge (since it's not the same color as the fully transparent area - even fully transparent pixels have color information, even if you can't see it). There's no easy way to completely clean it out without plugins, but Pyrochild's Curves+ is one option. These settings will turn the slightly transparent parts into fully transparent ones: alpha channel, draw straight lines, move the bottom left control point rightwards. Important parts highlighted in red. The resulting edge detect in the bottom right corner:

As Tactilis explained above, the component that handles some file types (artificially) limits the width/height even if the file format itself supports it. As a workaround, I checked that it's possible to save a .tiff or a .bmp beyond 65535 pixels, so you could do that and use some other tool that's capable of converting them to .png of that size - IrfanView for instance. .dds and paint.net native .pdn also work, but .dds is a lossy format and .pdn wouldn't be as likely to be supported by an external tool.

I use Panelling, since it's pixel-precise. It isn't lossless though since rotate/zoom works with floating point numbers and easily results in fractional pixels, and hence why plugins like Panelling are useful.

Control-V doesn't paste a layer, it pastes a selected object which will merge automatically once you deselect it. Paste a new layer with control-shift-V.

TR's Pixel Puller does something like this. There's also Pixel Pusher but it probably isn't quite as described.

There's TR's Custom Palette Matcher if you want to reduce to a specific set of colors, instead of having the colors automatically determined. It's very simple to use: make the palette you want using the paint.net palette swatch, save it, and by default those palettes are shown in the plugin's list. If your palette doesn't take up all the slots though you'll have to duplicate one of the colors to fill all the rest. There's also QuantDith, which supports dithering while lowering color count to a specific palette. It does support the same paint.net palette files but the plugin has a crash bug: you must edit the palette text files (click the "open palettes folder" on the palette swatch window to find them) and remove the 2 empty trailing lines manually that paint.net adds - else the plugin explodes on trying to load the custom palette.

I may have explained that poorly since I got kinda similar results as make color transparent, but as long as you found a resonably working solution it's all good.

Try Make Color Transparent from KrisVDM's pack. Choose black as the color to make transparent, 0.8 power, 0.0 full transparency threshold. The reason you get dim colors from using an alpha mask is because the colors themselves are already "tainted" by the black background, and applying an alpha mask won't remove that blackness in them, which brings up an alternative solution: using Curves+ from pyrochild's plugin pack. First grab the grayscale version of your picture to use for an alpha mask, then use Curves+ in HSV mode on the original. Uncheck "hue" and "saturation", and crank the only remaining curve (value) into a straight horizontal line at the top, i.e. you remove all value differences by maxing value out. Then apply the grayscale alpha mask on this value-removed image.

By saving the file as a JPG you already destroyed a lot of the small detail, so PNG was able to compress it higher than initially. This is not always the case, and you should just be saving JPG directly if the space the file takes is more important than quality. Both are unlayered formats so both save the same amount of pixels; there is no "rubbish" saved in either format. Your 10,000x10,000 image has 100,000,000 pixels taking up around 0.4 GB of memory, the image in memory has no relation to the file format; trying to move amounts of data like that is a big demand and blocky rendering (multiple threads working at their own pace on small blocks) is a much better option than sluggish, unresponsive attempts to move the entire huge image in one go.

Set palette to BW, adjust the brightness/contrast to get desired coverage of the gradient. Post-contrast is a bit weird and can't honestly figure out what it's doing, but pre/post-brightness both do about the same, and pre-contrast acts like contrast would usually. Also try the different dither patterns, ordered 2x2 might produce something very similar to what you had in your original images. Edit: Okay I guess I was late by a couple seconds for this, heh!

That, and back the last time I checked after the plugin was updated to better preserve colors, it still would oversaturate low-alpha pixels sometimes, but I've long forgotten how/deleted the files I got those results on. I just reinstalled the plugin to double check and am not seeing any of the above problems anymore (even colors in zero-alpha pixels are preserved) so I'll revise my post accordingly. Edit: looks like the file date on the apply mask plugin was shortly after I last checked in early-mid summer so I missed an update, my bad.

A couple options come to mind. Easy but chaotic and not very configurable: Alpha to Dots Less easy but configurable: extract the alpha mask, dither that with one of the several b/w dithering plugins, then replace the original alpha mask. Alpha extraction: Alpha to Gray from BoltBait's plugin pack Dither: BoltBait's pack has Floyd-Steinberg dithering, Halftone dithering from Ed Harvey's pack, QuantDith can do several styles of dithering, including ordered ones. Reapply alpha: BoltBait's pack has Apply Alpha Mask. There's this other Alpha Mask if you want to paste the mask from clipboard. Top left: original, top right: alpha to dots, bottom left: halftone, bottom right: QuantDith.

Assuming you're saving PNG files, they're not required to have any DPI information at all. Paint.net does add an optional header, but that header is metric and integer, stored as pixels per meter. The closest value to 96 DPI becomes 3778 pixels/meter and converting that back to DPI upon loading that file gets you 95.96. Since DPI is only an informational value for printing etc., it has absolutely no effect on the actual resolution of the image saved or loaded.

1. Probably, it's the everyday built-in go-to blur, no plugins required and works for most situations. 2. Only the plugin author could tell what they mean by True Blur, but they didn't really specify. 3. There's several flavors of Average Blur: ReMake's, Cookie's and KrisVDM's

And another quick v1.2 update, but while mulling over the idea of a levels tool working in HSLuv color space I managed to speed up the HSLuv library with some optimizations (the original code sure used a lot of heap allocations, they're about all gone now) and wanted to get the thing updated immediately and bundled the levels tool. Now it should no longer be noticeably slow under any circumstances, even on huge images.

How's something like this? Plugins: Inner Shadow, Adjust Transparency from BoltBait's pack are the main ones. Duplicate your object that's on a transparent background. Run inner shadow on the topmost copy. Uncheck "keep background" in the options. Run adjustments->transparency. Check "ignore transparent pixels" and go full opaque (+1). This gets you a hard inner shadow already, but it's very crispy and pixelated. If you don't want to use other plugins, you can work in a higher resolution and downscale afterwards, but what I used here is Erode to withdraw the shadow slightly, then AA's Assistant from dpy's pack to soften it.

Oof, can't believe I managed to overlook a bug with saturation logic all this time, good thing I took another look before bed and got v1.1 out before too many people noticed...