Egregious....great word! Nice to hear someone with a good vocabulary...and CHDK too? LOL, I wore out a Powershot doing time lapse a few years ago - that was really fun. On your warnings about owning vs renting your OS and software: absolutely. You & Louis Rossman (sp?) would probably get along - Apple is just shameless about their repair policy, among other things! Cheers.
Thanks for the tutorial. I found this very helpful and I can't wait to give it a try. I just don't understand why everyone has to add a music bed behind an instructional video. Highly annoying.
I've published an updated version of this tutorial with less "dead air" and no music behind the talking. Thanks for suggesting it! ruclips.net/video/4fWlIDie-Ng/видео.html
G'Mic is available at gmic.eu/ and I don't know of anything that auto-aligns like Photoshop, other than some ancient Script-Fu plugins that I have not tried out. They really do need an auto-align feature.
Huh, the high iso noise reduction on every digital camera that I've owned in the last 20 years has been configurable, on, off, and usually some intermediate levels or something like a high. I'm wondering how this compares to using a tool like RegiStax? Have you tried it?
High ISO noise reduction (I'm assuming you really mean "long exposure noise reduction") in photography modes is actually a great thing to use because it's technically doing "dark frame subtraction." I'm not keen on the algorithmic noise reduction used in-camera for normal exposure, which relies on finding high-frequency components (stuff that is drastically different from what's immediately beside it) and smudging it out. In any case, noise reduction by taking a raw file into RawTherapee or similar tools will always do a better job than the NR done by the camera on the JPEG. Image stacking is fantastic, but it does require a still subject, tripod, and multiple exposures. I've only used Gimp to do it.
What we're doing with this layer blending stuff is reducing how far various forms of image noise deviate from what the true 100% clean image would be if no noise existed. We do this by taking multiple samples for each pixel and mixing them to get a better result. Now, why take the median and not the average? Take a set of pixel values; for simplification, we'll use single numbers. Let's say we have nine layers and the pixel being blended has the following values in each layer (this is the number set used, sorted for clarity): {87, 90, 104, 109, 110, 118, 120, 180, 226} The median is literally the middle value in the sorted set, so the value picked will be 110, easy peasy. The average is 1144/9 = 127. You can see in the set that most of the pixel samples have values closer to 110 than 127, but the three pixel samples that deviate heavily have skewed the results towards what we intuitively can see is not a good result. This is why taking the median works so much better: we're trying to pull the signal out of the noise and we don't want to add random spikes in noise back into the result. If you set layer opacity to (100 / layer_count) percent on every layer, you'll get roughly the same effect as averaging. The reason you can get away with doing medians on a few layers at a time is that they will still trend toward the median. If we split the set above into the randomly chosen {87, 118, 180} {90, 104, 120} {109, 110, 226} and median each, you get three new layers which make a set {104, 110, 118} which then has a median of 110. There are arrangements where this will not get to 110 (any set combination where 110 isn't the median of at least one of them) but even if we try to put the worst-case scenario together we'll still end up with something closer to 110 than if we averaged everything together. An example of an intentionally bad case: {87, 90, 104} {109, 118, 120} {110, 120, 226} -> {90, 118, 120} -> 118. Still closer to 110 than averaging would ever get. As you'd expect, more layers (more samples per pixel) will give more accurate results.
Wow, amazing and quick support!👍 I got the difference and see now that it's usually better to use median, cause on one layer the pixel might have gone crazy. Of course there are theoretical cases where average is better (82,83,84,84,130,130,132), but yea...
David B Those theoretical cases are extremely unlikely to occur for image noise. Noise is generally not a regular thing, and even if some pixels fell into that kind of unlikely data set, the majority simply won't do that.
great video! only question I have is what happens if you have one layer that has slightly moved from the other frames- is there a way to align the layers then blend?
Unfortunately, there is no tool in Gimp or G'MIC that works to auto-align layers. G'MIC has a layer alignment item but for whatever reason it simply doesn't seem to work. I do have a clever trick for hand-aligning layers that are exposed nearly identically only horizontally/vertically shifted and not rotated though. With at least two layers open and active, make the top layer's opacity exactly 50% (opacity is right above the layer list) and then invert the colors in that top layer. You'll notice that the inverted top and non-inverted bottom layers mixing creates an effect not unlike the classic embossing effect, with a layer of middle grey color and any "edges" appearing to stick out of or crush into the image. Zoom in to 200% or more and slowly move one of the layers until you see the emboss effect mostly disappear. If the whole image now looks almost entirely grey (no obvious big "edges" showing) when you zoom out, they're now aligned enough to stack. Reverse the top layer effects (go to 100% opacity and invert it again) and you now have two aligned layers. If you do this to the top layer and then activate and move the lower layers one by one from the bottom up, you'll roughly align every layer to match the position of the top one. Keep in mind that any layer alignment will result in artifacts around the edges after applying effects across the layers, so you'll want to crop off the edges once you finish up.
You must set the G'MIC output to output to a new layer. Look for the four white drop-down boxes grouped together, usually near the bottom-right of the window. Make those match what you see in the video.
Great demonstration! I think you mentioned it at the end, ‘making adjustments prior to combining when using the RAW files’. My question is/was when to adjust gamma, saturation, color temp, contrast, etc., before or after combining layers/frames? Also, what computer and memory gave you the memory error issue?
I had a custom built desktop with 16GB of RAM; the problem is that DSLR photos are massive when uncompressed and the intermediates created by G'MIC are even worse. It's not a hardware issue, in any case; it's a software issue of some sort. If you're going to adjust the images during RAW processing, it's important that you either apply the exact same settings to all files OR you tweak them to match the appearance of the other exposures as closely as possible, depending on if the lighting changed at all.
Ideally there is no difference other than noise. The stacking shown here is using multiple images of the exact same thing to reduce the visible noise. I'm not sure that there's a specific term to differentiate it from focus stacking, but if I had to invent one, I'd call it "noise stacking" or "clarity stacking" or perhaps "sharpness stacking," since image noise has the effect of reducing overall sharpness and clarity. Another kind of "stacking" that's not usually referred to using the term "stacking" is high dynamic range (HDR) photography, which I have a tutorial video for as well. In HDR, you take multiple photos at different levels of exposure (usually -1, 0, +1 EV, also called "exposure bracketing") and you merge the photos together using selective masking of the over- and under-exposed areas in the center exposure to reduce the overall image contrast. HDR lets you take a photo of a white flower on an apple tree with sunlight hitting it (as in my HDR tutorial) and recover the texture of the flower petals from an under-exposed picture, which the sunlight otherwise just blows out to a solid white area with no detail at all. Focus stacking is where you algorithmically cut out the portions of each image that are the most in focus (i.e. the sharpest) and layer those in-focus zones together into a single image with far greater focus depth than is technically possible to achieve in a single photo. Focus stacking is nearly always related to macro photography, since the extremely small objects being photographed and the extremely short focal distances required to shoot them tend to have extremely narrow focus depth as a result. Those cool pics of a fly's eye where the whole eye is in focus were likely a set of several images taken at very slightly different focus targets, then combined using focus stacking.
@@JodyBruchon Thank you for the quick and detailed reply! So you say that for the kind of stacking you showed in this video you just have to shoot few images of the same thing with no change in the camera settings?
Yes, the stacking being done here requires several identical photos, so the camera needs to be on a tripod, all camera settings need to be locked to manual settings (M mode, ISO not set to Auto), the things in the shot should not move, and several photos should be shot in rapid succession. You'll usually want the continuous shooting feature turned on for this if you are using all but the cheapest point-and-shoots.
What happens if there's a wind moving some of the branches? Is GIMP smart enough to ignore those pictures or does this only work under windless conditions.
No, you would need to exclude those yourself. However, if you have enough layers, it's possible that the movement will cancel out. If in doubt, you could manually use a black pencil tool on the areas of concern on the layers that have issues; the black should be removed by the median function.
It will work if you correct the exposure back, and may work better if use the techniques in my HDR photos with GIMP video to mask the layers before correcting the exposure. The only bad thing is that correcting exposure will also bring out hidden JPEG artifacts more, so if you can, develop RAW files to a format that supports 16 bit per channel color (PNG/TIFF) and import that instead of JPEGs. You should also use GIMP 2.10 and make sure that you edit the image in 16 bpc color mode instead of 8 bpc.
@@JodyBruchon I mostly use photos from my drone that doesn’t produce raw photos unfortunately. Right now I use software that automatically takes three photos one over, one under and one correctly exposed. I believe jpg photos is what it produces. Am I on the right track?
@@brandonwarren8351 Assuming you can get them perfectly aligned, you can exposure-correct the +/- photos to match the normal one and then use the Blend Layers (Median) to merge them into a lower-noise layer. You have two major problems: you need to match the exposure as closely as possible and you need to align the layers almost perfectly. GIMP has no facility for automatic alignment and I don't think there's an inter-layer exposure matching function either, but look up my video about "aligning layers by hand in GIMP." If your layers are already aligned (the drone was quite stable and took the shots very quickly, for example) then you can easily fix the exposure by making the top layer 50% opacity, inverting the layer's colors, and adjusting the brightness until the whole image looks like good old middle grey. Then, invert the layer again and you now have two very closely exposure-matched layers to median blend.
Ouch...yeah, G'MIC recently changed their installation method and it requires an extra step. I probably should do a follow-up video. The missing step is that once you've installed G'MIC, you have to add the path to the plugin to Gimp's plugin path. It's not trivial to explain by text. If you are on Windows, it's under your user account and the full path will be like "C:\Users\username\.gimp-2.8\plug-ins\gmic_gimp_qt". It'd be easier to explain in a video but I can't do it right now.
@@nevinjohnson1853 No; it's different depending on what OS you are using, but the gist of it is that you need to download G'MIC and unpack it somewhere, then add the folder you unpacked it into to the Gimp plugin folder list in the Gimp preferences.
I took some pictures of the lunar eclipse, but the moons aren´t aligned due to the moon´s own movement. The align function of G´MIC can´t cope with that at all for some reason. Is there another way?
Turn visibility for all but the bottom 2 layers off. Set the top visible layer to 50% opacity (at the top of the Layers panel). Zoom in on the moon. Align as best as possible by hand. Set the top visible layer back to 100% opacity and turn off its visibility. Turn on visibility for the next layer up and repeat the alignment procedure. It sucks but it's really the only other way to do it that I know of.
I tried to stack 2 pictures with different focus points on the subject. I used a tripod. It never aligns correctly and the result is just a blurred image.
It won't help with rotational problems, but if the misalignment is a simple up-down/left-right shift from a button press or something of that nature, you might be able to align them with my video "Aligning Layers By Hand in GIMP" - ruclips.net/video/Jt8hvqzqcdI/видео.html
So I'll ask about my other huge peeve with modern gimp... How do we change the upper range of the brush size slider? 1,000 is way too much for my workflow.
I have never dealt with that issue, but you'll notice that the pointer changes between the top half and bottom half of the slider. The "up arrow" does coarse drag adjustments and the normal pointer does fine adjustments. I use the coarse adjust to get near the desired size and the fine adjust to further tweak the value to suit. It's not what you're asking for but maybe it's a good enough workaround.
So basically you took 18 pictures with the exact same settings? (ie. as if you pressed the shutter button 18 times without adjusting the camera each time?)
MrCkl88 I took 18 pictures in one continuous sequence without moving the camera or changing anything. A tripod or other solid camera mount is mandatory. The best way to minimize the shaking (other than the tripod) is to use a shutter remote if your camera supports one so you can press the button without touching the camera and tripod. Even the slightest movement between frames will make the process more difficult since you'll have to either align or discard the layers that are different. The align layers function in G'MIC doesn't seem to work very well and eyeballing the alignment doesn't always work.
I generally do not shoot RAW since it is often a waste of time and disk to do so. If it was something extremely important I'd shoot RAW + JPEG but I find that JPEG is more than good enough in nearly all of my shooting circumstances.
I'm sorry to hear that. It seems that there are extremely few GIMP developers that use macOS. Given that Apple is such a customer-hostile company and their products are massively overpriced, I am not surprised that several "computer people" have shied away from them in general.
Nice video. Good point about the camera's own denoising. Two quibbles, though: - At 5:53 when you choose which images to blend as a second batch, it appears you selected the result of the first batch, to blend into the second batch. It seems to me that the right thing to do would be to do three batches independently (with no inclusion of prior result layers in subsequent batches), *then* do a fourth blend of the previous three result layers. (At 6:35, you even briefly describe doing exactly that.) As it is in the video example, however, the layers that contributed to the first blend got very little influence on the median calculation for the value for each pixel in the final result layer, since the layers of the first batch were collectively reduced to having just one layer's weight in that final comparison. - I found the tense, repetitive music slightly distracting, but maybe that's just me. Good video overall though, and kudos for rolling with the error and without hesitation taking it as an opportunity to give a further useful tip. (Hmm... Or was there a previous recording where you stopped recording at that point and tested closing other programs or checked just how many layers needed to be disabled... and then re-recorded actually knowing you'd get the error? I wouldn't blame you if that was the case. That said, my assumption is that the error really was unexpected.)
The music choice was not the greatest. This video was made with a lot less experience under my belt, so I made a lot of mistakes. To clarify the median thing: median calculations don't have "weight." Median takes the middle value in a set (or the interpolated middle in the case of a set with an even count of values), so taking intermediate medians of subsets and then taking the median of those medians should result in a final number that is not necessarily identical to the median for the full set, but isn't too far off either. Image noise is mostly photon randomness rather than read noise in the majority of modern digital cameras, so it is sufficiently random that a random subset should also be random. True randomness tends to have even distribution once enough samples are taken (and that's all the pixels of the layers are--samples of light levels) so subsets of a larger set will tend to have even distribution as well. It doesn't matter how you slice and dice the data all that much as long as you have enough samples for each median you take. Including the first set result in the second set pulls in the median of the first set as a data point, and presumably that data point is more likely to be the correct value for that pixel (by virtue of being a median taken from several samples already) that its inclusion will make the result of the next set more accurate. For example, if the sorted main set of numbers is {0,2,2,4,5,6,6,9,9} but is randomly arriving as {9,5,2,4,6,9,0,6,2} and we take medians three at a time using my method and then yours, we get: {9,5,2} = 5, {5,4,6} = 5, {5,9,0} = 5, {5,6,2} = 5 {9,5,2} = 5, {4,6,9} = 4, {0,6,2} = 2, {5,4,2} = 5 In both cases we achieve the same median. It should also be clear that if the noise were not random (as if we used the sorted numbers), my way would not work because it would result in 9 instead of 5, and your way would achieve the correct result. Thus, you are right if the data were not sufficiently random.
I didn't catch the end of the comment, sorry. Thank you for the kind words. I filmed this in one go and I didn't expect the error. I did do the stacking before I took the video and didn't run into that error. I figured that anyone who ran into the error might want to know how to get around it, so I figured it out. "Couldn't allocate memory" and similar are kind of obvious as to what's going wrong, so I figured less layers at once would work. Uncompressed images are amazingly large, and a load of those plus intermediate processing data can balloon to insane amounts of memory usage.
There's a very good chance that security software hasn't seen it very often because it's not a popular package like Gimp itself is. It's probably a false positive. Submit the "Trojan" to VirusTotal and see what comes back.
*WATCH THE 2024 UPDATED VERSION, NOT THIS ONE:* ruclips.net/video/4fWlIDie-Ng/видео.html
not gonna lie, I'm shit at art, I don't even like making it, it goes against my nature
but I tried this and WOW! I've actually done sth! Thanks man!
This has been added to my prestigious "photography" playlist.. LOL
"should not need to change them but..... huh" i feel your pain.
Egregious....great word! Nice to hear someone with a good vocabulary...and CHDK too? LOL, I wore out a Powershot doing time lapse a few years ago - that was really fun.
On your warnings about owning vs renting your OS and software: absolutely. You & Louis Rossman (sp?) would probably get along - Apple is just shameless about their repair policy, among other things!
Cheers.
Thanks for the tutorial. I found this very helpful and I can't wait to give it a try. I just don't understand why everyone has to add a music bed behind an instructional video. Highly annoying.
I made this many years ago. I wouldn't put music there today.
I've published an updated version of this tutorial with less "dead air" and no music behind the talking. Thanks for suggesting it! ruclips.net/video/4fWlIDie-Ng/видео.html
Thanks a lot for showing!
For Ubuntu Linux "sudo apt install gmic gimp-gmic" will install the packges.
Great video, thanks a lot for this.Hopefully I can install Gmic on my mac.
Maybe these install instructions will help: itstillworks.com/install-gmic-gimp-8504000.html
Thanks for the plugin, Its a pain on mac but it works.
👍
Hey Jody !
Do you know if there is a way to auto align layers like in photoshop, to avoid the focus breathing ? and how to get G'Mic-Qt plugin?
G'Mic is available at gmic.eu/ and I don't know of anything that auto-aligns like Photoshop, other than some ancient Script-Fu plugins that I have not tried out. They really do need an auto-align feature.
Huh, the high iso noise reduction on every digital camera that I've owned in the last 20 years has been configurable, on, off, and usually some intermediate levels or something like a high.
I'm wondering how this compares to using a tool like RegiStax? Have you tried it?
High ISO noise reduction (I'm assuming you really mean "long exposure noise reduction") in photography modes is actually a great thing to use because it's technically doing "dark frame subtraction." I'm not keen on the algorithmic noise reduction used in-camera for normal exposure, which relies on finding high-frequency components (stuff that is drastically different from what's immediately beside it) and smudging it out. In any case, noise reduction by taking a raw file into RawTherapee or similar tools will always do a better job than the NR done by the camera on the JPEG.
Image stacking is fantastic, but it does require a still subject, tripod, and multiple exposures. I've only used Gimp to do it.
What´s the difference to using "Blend (average all)"? And why is Blend Median better? Great video by the way!
What we're doing with this layer blending stuff is reducing how far various forms of image noise deviate from what the true 100% clean image would be if no noise existed. We do this by taking multiple samples for each pixel and mixing them to get a better result. Now, why take the median and not the average?
Take a set of pixel values; for simplification, we'll use single numbers. Let's say we have nine layers and the pixel being blended has the following values in each layer (this is the number set used, sorted for clarity): {87, 90, 104, 109, 110, 118, 120, 180, 226} The median is literally the middle value in the sorted set, so the value picked will be 110, easy peasy. The average is 1144/9 = 127. You can see in the set that most of the pixel samples have values closer to 110 than 127, but the three pixel samples that deviate heavily have skewed the results towards what we intuitively can see is not a good result. This is why taking the median works so much better: we're trying to pull the signal out of the noise and we don't want to add random spikes in noise back into the result. If you set layer opacity to (100 / layer_count) percent on every layer, you'll get roughly the same effect as averaging.
The reason you can get away with doing medians on a few layers at a time is that they will still trend toward the median. If we split the set above into the randomly chosen {87, 118, 180} {90, 104, 120} {109, 110, 226} and median each, you get three new layers which make a set {104, 110, 118} which then has a median of 110. There are arrangements where this will not get to 110 (any set combination where 110 isn't the median of at least one of them) but even if we try to put the worst-case scenario together we'll still end up with something closer to 110 than if we averaged everything together. An example of an intentionally bad case: {87, 90, 104} {109, 118, 120} {110, 120, 226} -> {90, 118, 120} -> 118. Still closer to 110 than averaging would ever get.
As you'd expect, more layers (more samples per pixel) will give more accurate results.
Wow, amazing and quick support!👍
I got the difference and see now that it's usually better to use median, cause on one layer the pixel might have gone crazy.
Of course there are theoretical cases where average is better (82,83,84,84,130,130,132), but yea...
David B Those theoretical cases are extremely unlikely to occur for image noise. Noise is generally not a regular thing, and even if some pixels fell into that kind of unlikely data set, the majority simply won't do that.
Jody Bruchon Exactly, that's what I noticed, too.
great video! only question I have is what happens if you have one layer that has slightly moved from the other frames- is there a way to align the layers then blend?
Unfortunately, there is no tool in Gimp or G'MIC that works to auto-align layers. G'MIC has a layer alignment item but for whatever reason it simply doesn't seem to work. I do have a clever trick for hand-aligning layers that are exposed nearly identically only horizontally/vertically shifted and not rotated though.
With at least two layers open and active, make the top layer's opacity exactly 50% (opacity is right above the layer list) and then invert the colors in that top layer. You'll notice that the inverted top and non-inverted bottom layers mixing creates an effect not unlike the classic embossing effect, with a layer of middle grey color and any "edges" appearing to stick out of or crush into the image.
Zoom in to 200% or more and slowly move one of the layers until you see the emboss effect mostly disappear. If the whole image now looks almost entirely grey (no obvious big "edges" showing) when you zoom out, they're now aligned enough to stack. Reverse the top layer effects (go to 100% opacity and invert it again) and you now have two aligned layers.
If you do this to the top layer and then activate and move the lower layers one by one from the bottom up, you'll roughly align every layer to match the position of the top one.
Keep in mind that any layer alignment will result in artifacts around the edges after applying effects across the layers, so you'll want to crop off the edges once you finish up.
Do you have a step by step list anywhere? Great video!
No, but I will make one for you tonight.
Thank you Jody! Where will I be able to find the list?
Hi Jody, have you been able to create a step by step lest? Thanks!
kelsi6273 Not yet. Give me a bit.
kelsi6273 List is now in the description.
It doesn't matter which layer is selected when using this filter in the plugin?
Correct.
did as guided by the video, but does not make a new layer.. anything i did wrong?
You must set the G'MIC output to output to a new layer. Look for the four white drop-down boxes grouped together, usually near the bottom-right of the window. Make those match what you see in the video.
Great demonstration! I think you mentioned it at the end, ‘making adjustments prior to combining when using the RAW files’. My question is/was when to adjust gamma, saturation, color temp, contrast, etc., before or after combining layers/frames? Also, what computer and memory gave you the memory error issue?
I had a custom built desktop with 16GB of RAM; the problem is that DSLR photos are massive when uncompressed and the intermediates created by G'MIC are even worse. It's not a hardware issue, in any case; it's a software issue of some sort. If you're going to adjust the images during RAW processing, it's important that you either apply the exact same settings to all files OR you tweak them to match the appearance of the other exposures as closely as possible, depending on if the lighting changed at all.
Owen Wilson?
What is the difference between the images that you are stacking? Is it the focus or something else?
Ideally there is no difference other than noise. The stacking shown here is using multiple images of the exact same thing to reduce the visible noise. I'm not sure that there's a specific term to differentiate it from focus stacking, but if I had to invent one, I'd call it "noise stacking" or "clarity stacking" or perhaps "sharpness stacking," since image noise has the effect of reducing overall sharpness and clarity.
Another kind of "stacking" that's not usually referred to using the term "stacking" is high dynamic range (HDR) photography, which I have a tutorial video for as well. In HDR, you take multiple photos at different levels of exposure (usually -1, 0, +1 EV, also called "exposure bracketing") and you merge the photos together using selective masking of the over- and under-exposed areas in the center exposure to reduce the overall image contrast. HDR lets you take a photo of a white flower on an apple tree with sunlight hitting it (as in my HDR tutorial) and recover the texture of the flower petals from an under-exposed picture, which the sunlight otherwise just blows out to a solid white area with no detail at all.
Focus stacking is where you algorithmically cut out the portions of each image that are the most in focus (i.e. the sharpest) and layer those in-focus zones together into a single image with far greater focus depth than is technically possible to achieve in a single photo. Focus stacking is nearly always related to macro photography, since the extremely small objects being photographed and the extremely short focal distances required to shoot them tend to have extremely narrow focus depth as a result. Those cool pics of a fly's eye where the whole eye is in focus were likely a set of several images taken at very slightly different focus targets, then combined using focus stacking.
@@JodyBruchon Thank you for the quick and detailed reply! So you say that for the kind of stacking you showed in this video you just have to shoot few images of the same thing with no change in the camera settings?
Yes, the stacking being done here requires several identical photos, so the camera needs to be on a tripod, all camera settings need to be locked to manual settings (M mode, ISO not set to Auto), the things in the shot should not move, and several photos should be shot in rapid succession. You'll usually want the continuous shooting feature turned on for this if you are using all but the cheapest point-and-shoots.
@@JodyBruchon Thank you so much for responding!
I noticed you blended layers without resizing the images, maybe you can get more sub-pixel detail by resizing it????
What happens if there's a wind moving some of the branches? Is GIMP smart enough to ignore those pictures or does this only work under windless conditions.
No, you would need to exclude those yourself. However, if you have enough layers, it's possible that the movement will cancel out. If in doubt, you could manually use a black pencil tool on the areas of concern on the layers that have issues; the black should be removed by the median function.
i believe its Pronounced "Gimic"
I can't open Gmic
Will this work for aeb bracketed photos?
It will work if you correct the exposure back, and may work better if use the techniques in my HDR photos with GIMP video to mask the layers before correcting the exposure. The only bad thing is that correcting exposure will also bring out hidden JPEG artifacts more, so if you can, develop RAW files to a format that supports 16 bit per channel color (PNG/TIFF) and import that instead of JPEGs. You should also use GIMP 2.10 and make sure that you edit the image in 16 bpc color mode instead of 8 bpc.
@@JodyBruchon I mostly use photos from my drone that doesn’t produce raw photos unfortunately. Right now I use software that automatically takes three photos one over, one under and one correctly exposed. I believe jpg photos is what it produces. Am I on the right track?
@@brandonwarren8351 Assuming you can get them perfectly aligned, you can exposure-correct the +/- photos to match the normal one and then use the Blend Layers (Median) to merge them into a lower-noise layer. You have two major problems: you need to match the exposure as closely as possible and you need to align the layers almost perfectly. GIMP has no facility for automatic alignment and I don't think there's an inter-layer exposure matching function either, but look up my video about "aligning layers by hand in GIMP." If your layers are already aligned (the drone was quite stable and took the shots very quickly, for example) then you can easily fix the exposure by making the top layer 50% opacity, inverting the layer's colors, and adjusting the brightness until the whole image looks like good old middle grey. Then, invert the layer again and you now have two very closely exposure-matched layers to median blend.
Can't install it
hi you showed us a little in how to set up gimp thank you how about showing us how to install gmic ? thank you again
Ouch...yeah, G'MIC recently changed their installation method and it requires an extra step. I probably should do a follow-up video. The missing step is that once you've installed G'MIC, you have to add the path to the plugin to Gimp's plugin path. It's not trivial to explain by text. If you are on Windows, it's under your user account and the full path will be like "C:\Users\username\.gimp-2.8\plug-ins\gmic_gimp_qt". It'd be easier to explain in a video but I can't do it right now.
hi thank you i ham a ubuntu user @@JodyBruchon
@@JodyBruchon have you made a video for this? i am having trouble figuring this out.
@@nevinjohnson1853 No; it's different depending on what OS you are using, but the gist of it is that you need to download G'MIC and unpack it somewhere, then add the folder you unpacked it into to the Gimp plugin folder list in the Gimp preferences.
I took some pictures of the lunar eclipse, but the moons aren´t aligned due to the moon´s own movement. The align function of G´MIC can´t cope with that at all for some reason. Is there another way?
Turn visibility for all but the bottom 2 layers off. Set the top visible layer to 50% opacity (at the top of the Layers panel). Zoom in on the moon. Align as best as possible by hand. Set the top visible layer back to 100% opacity and turn off its visibility. Turn on visibility for the next layer up and repeat the alignment procedure. It sucks but it's really the only other way to do it that I know of.
@@JodyBruchon Thanks! The remaining problem for me is just that I don't know how to drag the layers :D
David B Use the Move tool, it looks like a cross with arrow points in the Toolbox on the left.
@@JodyBruchon Thank you very much!
does it support nikon z6 raw files .NEF ??
I don't use a Nikon Z6 so I don't know, but I assume so.
Could it be the memory sucking up some memory?
I tried to stack 2 pictures with different focus points on the subject. I used a tripod. It never aligns correctly and the result is just a blurred image.
It won't help with rotational problems, but if the misalignment is a simple up-down/left-right shift from a button press or something of that nature, you might be able to align them with my video "Aligning Layers By Hand in GIMP" - ruclips.net/video/Jt8hvqzqcdI/видео.html
Were the pics RAW or JPG when you pulled them into GIMP?
thanks keep up the good work.
JPEG. Gimp can't open raw files directly.
Floating tool boxes were a horrid idea. Thanks for telling me how to turn them off!
I couldn't agree more, and I'm happy to have helped!
So I'll ask about my other huge peeve with modern gimp... How do we change the upper range of the brush size slider? 1,000 is way too much for my workflow.
I have never dealt with that issue, but you'll notice that the pointer changes between the top half and bottom half of the slider. The "up arrow" does coarse drag adjustments and the normal pointer does fine adjustments. I use the coarse adjust to get near the desired size and the fine adjust to further tweak the value to suit. It's not what you're asking for but maybe it's a good enough workaround.
So basically you took 18 pictures with the exact same settings? (ie. as if you pressed the shutter button 18 times without adjusting the camera each time?)
MrCkl88 I took 18 pictures in one continuous sequence without moving the camera or changing anything. A tripod or other solid camera mount is mandatory. The best way to minimize the shaking (other than the tripod) is to use a shutter remote if your camera supports one so you can press the button without touching the camera and tripod. Even the slightest movement between frames will make the process more difficult since you'll have to either align or discard the layers that are different. The align layers function in G'MIC doesn't seem to work very well and eyeballing the alignment doesn't always work.
Thanks for the info. I didn't realize you could remove noise like this. Pretty neat.
Does it support tiff or dng ?
TIFF yes. DNG you'll need to use something like UFRaw or RawTherapee to bring in.
@@JodyBruchon thanks :)
Great tutorial. Did you shoot in RAW? I noticed that the pictures are in jpeg. Thanks.
I generally do not shoot RAW since it is often a waste of time and disk to do so. If it was something extremely important I'd shoot RAW + JPEG but I find that JPEG is more than good enough in nearly all of my shooting circumstances.
Loved the tutorial! Can you make a video on autoRetouch? It is an AI powered photo editing platform. It is offering a one month free trial as well!
G'Mic-Qt for MAC OS 10 ???
Looks like you can get what you're after on this guy's page: www.partha.com
Frustration Alert: As of September 2021 G'MIC-Qt is not available for macOS.
I'm sorry to hear that. It seems that there are extremely few GIMP developers that use macOS. Given that Apple is such a customer-hostile company and their products are massively overpriced, I am not surprised that several "computer people" have shied away from them in general.
Anyone here know how to get this working on a Mac?
Try using McGimp instead of Gimp (it includes G'MIC): www.partha.com/
@@JodyBruchon thanks! Looks like the ticket!
I don't know why but every time I open or drag a image into gimp its small
Is it the zoom level?
I think it is
Nice video. Good point about the camera's own denoising. Two quibbles, though:
- At 5:53 when you choose which images to blend as a second batch, it appears you selected the result of the first batch, to blend into the second batch. It seems to me that the right thing to do would be to do three batches independently (with no inclusion of prior result layers in subsequent batches), *then* do a fourth blend of the previous three result layers. (At 6:35, you even briefly describe doing exactly that.) As it is in the video example, however, the layers that contributed to the first blend got very little influence on the median calculation for the value for each pixel in the final result layer, since the layers of the first batch were collectively reduced to having just one layer's weight in that final comparison.
- I found the tense, repetitive music slightly distracting, but maybe that's just me.
Good video overall though, and kudos for rolling with the error and without hesitation taking it as an opportunity to give a further useful tip. (Hmm... Or was there a previous recording where you stopped recording at that point and tested closing other programs or checked just how many layers needed to be disabled... and then re-recorded actually knowing you'd get the error? I wouldn't blame you if that was the case. That said, my assumption is that the error really was unexpected.)
The music choice was not the greatest. This video was made with a lot less experience under my belt, so I made a lot of mistakes.
To clarify the median thing: median calculations don't have "weight." Median takes the middle value in a set (or the interpolated middle in the case of a set with an even count of values), so taking intermediate medians of subsets and then taking the median of those medians should result in a final number that is not necessarily identical to the median for the full set, but isn't too far off either. Image noise is mostly photon randomness rather than read noise in the majority of modern digital cameras, so it is sufficiently random that a random subset should also be random. True randomness tends to have even distribution once enough samples are taken (and that's all the pixels of the layers are--samples of light levels) so subsets of a larger set will tend to have even distribution as well. It doesn't matter how you slice and dice the data all that much as long as you have enough samples for each median you take. Including the first set result in the second set pulls in the median of the first set as a data point, and presumably that data point is more likely to be the correct value for that pixel (by virtue of being a median taken from several samples already) that its inclusion will make the result of the next set more accurate.
For example, if the sorted main set of numbers is {0,2,2,4,5,6,6,9,9} but is randomly arriving as {9,5,2,4,6,9,0,6,2} and we take medians three at a time using my method and then yours, we get:
{9,5,2} = 5, {5,4,6} = 5, {5,9,0} = 5, {5,6,2} = 5
{9,5,2} = 5, {4,6,9} = 4, {0,6,2} = 2, {5,4,2} = 5
In both cases we achieve the same median. It should also be clear that if the noise were not random (as if we used the sorted numbers), my way would not work because it would result in 9 instead of 5, and your way would achieve the correct result. Thus, you are right if the data were not sufficiently random.
I didn't catch the end of the comment, sorry. Thank you for the kind words. I filmed this in one go and I didn't expect the error. I did do the stacking before I took the video and didn't run into that error. I figured that anyone who ran into the error might want to know how to get around it, so I figured it out. "Couldn't allocate memory" and similar are kind of obvious as to what's going wrong, so I figured less layers at once would work. Uncompressed images are amazingly large, and a load of those plus intermediate processing data can balloon to insane amounts of memory usage.
I don't even know how to open the first layer, let alone import a bunch of other layers.
You literally open one of the photos in the program.
Download G'Mic-Qt at your own risk. I tried and got a trojan warning.
There's a very good chance that security software hasn't seen it very often because it's not a popular package like Gimp itself is. It's probably a false positive. Submit the "Trojan" to VirusTotal and see what comes back.
I can't open gmic