Hey all,
So in some threads here a few of us were talking about stacking a bit more, versus just doing a really long exposure. It came up in light (get it?!) of the new Lee Super Stopper (15 stop ND filter), and the current existence of the Firecrest 16 stop ND filter. And of course there is stacking a 10 stop and 4 or 6 stop filter for similar stopping power. There are issues with really long exposure, especially on older sensors, and lots of room for error and problems. So another method is to use long exposure, but simply shorter exposures, to achieve a look normally associated with really long exposure. To make this relative statement have a more defined tone, take for example that 30 seconds while long exposure isn't that long, and you can achieve that in day light with a single 10 stop filter no problem (F22, ISO 100 with 10 stop filter) when shooting RAW without clipped highlights at high noon. But, what if you wanted the "look" of 2 minutes or 4 minutes or 8 minutes or more? You'd have to stack huge filters and do a single, uninterrupted, no mistake, hot sensor exposure for 8 minutes. I've done it at 8 minutes with a stack of filters and it was super noisy on a modern, newer sensor. So on my older sensors it would be a total mess.
What is stacking? Well there's stacking of ND filters, or there's digital stacking in Photoshop via layers. The result is very similar. Just a different approach. This experiment is based on shooting for long exposure looks (on the order of 2+ minutes), by doing several shorter exposures and stacking/blending them together digitally. The process is actually super easy, and doesn't require a bunch of technical masking, layering, brushing, etc. You literally just load them via File -> Script -> Load Files into a Stack (and check align & make into smart object). And then once they're loaded up into layers, you go to Layers -> Smart Objects -> Stack -> Mean. This takes the average basically and this is what takes your shorter exposures and blends them all together to get a single image that looks like a much longer exposure, equivalent to the total time. So four 30 second exposures would give you a 2 minute exposure, in theory, in terms of how it looks, but you wouldn't need the stacks of ND filters to achieve a 2 minute exposure in daylight, and you wouldn't need an intervalometer or anything, just do 4 consecutive shorter exposures, and stack them later. That's the idea at least.
Here's a link to a great article that really explains digital stacking well, with examples of screen shots from Photoshop so you can literally do it right now without a massive learning curve or anything.
The issues with stacking shorter exposures is that you run into a few problems with gaps between exposures. You still need a long exposure to work from, and several of them to stack them. The question is... how short of an exposure can you use, before issues arise with the blend? Especially regarding noise.
The purpose of this experiment is to find the threshold of two things: (1) how short of a sub can you do and still stack for a seamless blend, and (2) how to deal with gaps in time.
1. Dealing with a shorter sub, means taking more images. There's nothing inherently wrong with this. However, it does mean that there's potential to have more time between each sub, variable time, and that means you could end up with a lot of small gaps where information wasn't being gathered. This is a big deal if you consider some folk use "long exposure noise reduction" or "LENR" for short from now on. I do, because my sensor is old and very noisy (5D). I could use a newer sensor, but I don't, because I just like the look of the old 5D files, but this is my self-imposed handicap, someone else may not have to use LENR with shorter exposures, if the noise levels are low. The noise gets worse as the exposure goes longer, due to heat. So doing a single 2 minute exposure is significantly more noisy than a single 30 second exposure. But, again, the main issue here is actually dealing with exposure times and exposure numbers, with gaps, and how that might effect the final image when stacking.
2. Gaps in time are a reality. I used LENR with each sub I made. I really stress it, I left LENR on and did shorter subs, so that the increments of time that I was "down" (doing the LENR processing) was less, so the gaps would be smaller. Had I done longer exposures, the LENR would have been longer and the gaps would have been larger and more pronounced. Ideally, you do this without using LENR. It's that simple. Also, to minimize gaps, you should have no down time. Intervalometer helps with this. Longer single exposures will stack to look more like a natural single super long exposure. However, this experiment is to test non-ideal conditions, or to show how bad it could be. We all have those days. Everything can look easy and magical in ideal conditions. So I wanted to mess around in the bad conditions and see how bad the results could be.
Experiment 1 :: Gaps & Gap Time ++++ Link: Experiment 2
In this first experiment, I went with a single 10 stop ND filter. It's a Haida PRO II Slim 10 stop filter. Cheap. Here's a quick link to my review of this filter: LINK. This is the filter I used in this first experiment series. Why a single 10 stop? Because it's a single filter. No filter stacks. No extraordinarily costly setup. A single, cheap, screw on filter that lets me use my lens hood and keeps the setup small, mobile and simple. 10 stops means I'm also very much limited in terms of how long of an exposure I can manage dependent on the light. Early morning and late evening, you can expose for a long time with a 10 stop. That's ideal. I wanted to stress unideal, so I went for a different time of day and different kind of light--the worst.
Conditions for Experiment 1:
1. Clouds are a must. I wanted something other than water to blur.
2. High noon or close to it. The reason: worst light possible to expose in. Also, brightest light, so it really stresses shorter exposure values.
3. Getting very long exposure, ie, 2 minutes minimum as the final exposure time, using shorter subs.
4. Exposure values with a single 10 stop filter were: 10 seconds, F16, ISO 100 (*).
5. Long Exposure Noise Reduction is on. This will cause me to have forced gaps!
6. There must be gaps, to stress the importance of exposure time, versus stress of potential noise.
(*) I could have used F22 and ISO 50 to get 40 second exposures, but I didn't because I find F22 to be soft and not worth it, and ISO 50 is not a true ISO value, so kind of pointless, and would be better to just adjust exposure down in RAW for the same effect, without the loss or extrapolation of data. Due to this, I was limited to 10 seconds of exposure time in the non-ideal conditions stated above. To go longer, I would have needed to add a 2 stop filter to hit 40 seconds. Or a single 1 stop filter or CPL filter, to achieve 30 seconds with a little overhead. I chose not to stack, because the point is to avoid that in the first place.
I did 4 experimental images to get a feel for this and to get some observable results.
Image 1
Exposure value 10 seconds, F16, ISO 100
10 stop Haida Slim PRO II ND filter
Number of subs: 14
Total exposure time: 140 seconds (10 seconds x 14)
LENR time: 10 seconds (10 second gaps between all images!). Total LENR time: 140 seconds.
Straight out of camera sub with zero editing:
BridgeStackA_01_SingleSub by Martin Wise, on FlickrDigitally stacked 14 subs and edited (final image):
BridgeStackA_01_Clr by Martin Wise, on FlickrNotes: I think the gaps are pretty obvious. It looks like a repeating pattern almost was brushed in there. I do note the water blended nicely and is blurred nicely but it was already fairly calm to begin with. The 10 second gaps from LENR really show up here.
Image 2
Exposure value 10 seconds, F16, ISO 100
10 stop Haida Slim PRO II ND filter
Number of subs: 12
Total exposure time: 120 seconds (10 seconds x 12)
LENR time: 10 seconds (10 second gaps between all images!). Total LENR time: 120 seconds.
Straight out of camera sub with zero editing:
BridgeStackA_02_SingleSub by Martin Wise, on FlickrDigitally stacked 12 subs and edited (final image):
BridgeStackA_02_Clr by Martin Wise, on FlickrNotes: I think the gaps are still quite noticeable (on the right). However, it looks a little smoother (I think maybe due to the clouds moving a bit faster across the frame in this case).
Image 3
Exposure value 10 seconds, F16, ISO 100
10 stop Haida Slim PRO II ND filter
Number of subs: 13
Total exposure time: 130 seconds (10 seconds x 13)
LENR time: 10 seconds (10 second gaps between all images!). Total LENR time: 130 seconds.
Straight out of camera sub with zero editing:
BridgeStackA_03_SingleSub by Martin Wise, on FlickrDigitally stacked 13 subs and edited (final image):
BridgeStackA_03_Clr by Martin Wise, on FlickrNotes: Again, the gaps are noticeable, but the pattern is throughout so it almost looks ok. But still obvious gaps.
Image 4
Exposure value 10 seconds, F16, ISO 100
10 stop Haida Slim PRO II ND filter
Number of subs: 21
Total exposure time: 210 seconds (10 seconds x 21)
LENR time: 10 seconds (10 second gaps between all images!). Total LENR time: 210 seconds.
Straight out of camera sub with zero editing:
BridgeStackA_04_SingleSub by Martin Wise, on FlickrDigitally stacked 21 subs and edited (final image):
BridgeStackA_04_Clr by Martin Wise, on FlickrNotes: I went for a lot longer time, 210 seconds or 3.5 minutes. I wanted to see if the average would work better with shorter subs, if you simply collected more data. But, I think it doesn't matter. I still see the obvious gaps. It looks overall smoother to me, even with the gaps, and I think that's from the higher number of subs and more data over time, and it does simulate the kind of wispy clouds you'd expect to see in a 3+ minute exposure. However, I think the gaps are still long enough to be noticeable even with a higher amount of data to average out.
+++++++++++++++
Summary from Experiment 1.
I wanted to be able to do a single 2 minute exposure, just to compare to the digitally stacked subs results. However, I would have had to add a 2nd filter or use different exposure values (F22, ISo 50, etc). Ultimately I chose not to, because really, the first experiment is designed to stress gaps and gap time to see how they effect the stacked image. This is everyone's biggest concern with stacking, because it's the Achilles Heel of the technique and I think this first experiment really shows how obvious gaps can be. Sure, I could have turned off LENR and had minimal (milliseconds) of gap time total in an image if I were to take consecutive images with no LENR. But, I wanted to get some long gaps to keep LENR (as I know many of us use it on noisy sensors), and to see how bad this unideal condition would be for the final image.
10 second gaps are too long and show up, even with nearly double the exposure time to get a higher mean of data. The gaps need to be shorter. Much shorter.
Long exposure noise reduction (LENR) really is a problem with this method of long exposure. It forces gaps. It's gap time is equivalent to the exposure time, so it takes twice as long to do the sub, and you lose as much exposure time as you had in exposure time. Again, this was on purpose, to test this as an unideal condition. And this one to me is probably the biggest issue with this method. Gaps definitely hurt the image and show up obviously.
Time of day was obviously bad. But this was on purpose, to stress the technique in bad conditions to see results. This forced me to shorter sub exposures. However, I think with even a 30 second exposure and 30 second gap, the gaps would be more pronounced, due to missing 30 seconds of data collection, multiple times (4 times, instead of 12 times). I will test this on the next experiment, but based on these results, I have a strong prediction that the 30 second gaps will be even more obvious.
Next Experiments:
I will test the following concepts (and will include more, if anyone has suggestions or ideas) on the next experiments as I get time.
(A) Longer subs (30 seconds to 120 seconds each), with LENR on (to get a bigger gap). Purpose: to see if the gap is worse with more time, even with longer subs.
(B) Short subs (10 seconds) and longer subs (30 seconds to 120 seconds) with LENR OFF to have next to no gaps, to see the difference in quality of stack between short & long subs (and compare noise).
(C) Digital stack of any method above, compared to a physical ND filter stack resulting in the same or very close exposure values and exposure time, to compare the digital stack to a single long exposure from high stopping power filter use (will require 2 filters minimum).
I'm about to start my work week, so it will be a few days. I'll have time to think on it, and get ideas from everyone in the mean (get it!??) time.
Very best,
