On the Photography StackExchange site, I came across an interesting question: “Factoring out all features and just considering the Quality of the light, What is the difference between a very expensive flash and a average priced flash?”
I was aware that people had posted opinions on general quality of manufacturing and durability in a professional-use situation and product lifetime, and I had decided to get one of each (one expensive, one cheap). The only feature difference that I use is the secondary lamp on the high-end brand. Since I have them, I thought I could easily compare the results.
The Test Shots
I shot the same picture using a Yongnuo YN568EXII (currently sold for $105) and a Metz 58AF2 ($400 for those versions still for sale). I used a tripod with the flash mounted on-camera, and fully specified the exposure as ISO 100, 1/250 second at f/22. The flashes were set to use ETTL to determine the light output automatically, with the zoom head set to 70mm, and the secondary lamp on the Metz disabled.
The raw files can be downloaded via Dropbox. IMG_4236 was taken using the Yongnuo, and IMG_4237 with the Metz.
RAW Data Analysis
To my surprise, the immediate appearance of the histogram on the camera showed striking differences! The Metz showed a substantially broader range of tonal values, while the Yongnuo showed a narrow spike dominating the display.
To get a closer look, I used RawDigger, which will give information about the real recorded values in the RAW file without any kind of interpretation going on.
Where are these differences coming from? I’ll report only the green values because the shapes of the histogram are the same in all channels.
For middle grey, the background border areas of the card, Yongnuo (Y) shows an average value of about 702 and σ (standard deviation) is 32. For Metz (M), the average is 852 with σ 25. The darkest and lightest (un-numbered) patches are 68–3013, while for M they range from 55½–2417, with similar variances on each brand.
For M, the brightest un-colored base paper is not the brightest value recorded! Patches 13-A through 19-B, (C,D)-13 (cyans), and (C,D)-15 (yellows) are all brighter in the green channel. The brightest patches on M measure as 3142, where there is a visible spike at the right edge of the histogram. In Y, those patches are 2895, a little less than full bright, as expected.
This might be due to uneven flash coverage: I found the wall around the chart was also brighter towards the upper-right on the M, so the base-paper patch happens to lie in the corner that received the least light. On the other hand, the wall was more slightly brighter towards the top in the Y exposure. Since the camera was in the same position and the flash height is only about 1cm different, any effects due to angle should be the same on both.
Looking at the R,G,B values of patch L8, it is clear that the color balance is different: the Y has more blue than the M. In the histograms for the overall chart, the M shows double peaks in all three channels where the Y only has one. In the red channel, one peak between EV −2 and −3 (about sample value 350) while in M there is a double peak around this value; for green the situation is similar between EV −1 and −2; and for blue the peak just right of EV −2 in the Y corresponds to just the (smaller) right part of the double-peak on either side of EV −2, in the M. The blue channel also has another strong peak at EV −1 in the M that’s not present at all on Y. Clearly the colors are responding differently.
The overall exposure seems darker in the M photo just looking at what comes up in RawDigger, though range of values (as seen on the histograms) is the same on each, and the specific value recorded for middle gray is the other way around. For the wall to the right of the chart, the M is also a tiny amount brighter. But for the lower-left of the exposure, the wall is significantly darker in the M if you just look at the average.
The M exposure shows a lower average and a much lower σ, but the histogram pictures show fatter distributions on the M, and double peaks. This is confusing and not immediately obvious why, but RawDigger changes horizontal scales when moving between exposures, and the choice of scale is not intuitive. It appears that the exposure is about half a stop lower in this region for the M.
Without using careful measurements of the image pixel data, a side-by-side viewing appears to have a slight difference in exposure, but upon closer inspection only some things are exposed differently and the color balance is different. This definitely counts as a difference.
Developed Data Analysis
Loading the files into Lightroom, I synchronized changes to the two exposures. I cropped the chart so the histogram is dominated by neutrals and shows a clear peak that should be lined up in all channels. I also corrected the perspective so it’s roughly rectangular, which will simplify the later overlay comparison. Sharpening and Noise Reduction were both turned off.
The process used is 2012 Adobe Standard, and no adjustments where made to exposure.
The White Balance was set using the eyedrop tool over a section of grey background. The image is rather noisy with the color varying from pixel to pixel, visible in the loupe-zoom of the tool. So I tried a few spots to make sure it was representative, and verified that the RGB values of the neutral patches were showing the same in each channel.
The same WB was applied to both exposures, and the other exposure was visibly different, showing a color cast. Below are screen shots put together in a collage. You can see the histogram in LightRoom indicates that the neutral peaks don’t line up.
The Color Temperature of the two flashes are different, and neither matches the Flash preset color balance. The setting above, taken for the Yongnuo, is a temperature of 6650 with tint +10. The Metz is about 6072 with tint +14.
Next, I compare all the colored patches by cutting strips half as tall as the patches and alternating them from different exposures. Below, each patch, e.g. the yellow in L15, has the bottom of the box shown from the Yongnuo exposure, and the top of the box from the Metz. In addition, the left and right edges are painted in Photoshop as pure uniform grey.
After adjusting the WB individually for each exposure as explained above, it was clear from this rendering that the brightness varied across the chart, in different ways for each exposure! I used a (different) gradient exposure adjustment on each to approximate the uniform brightness of the grey background.
Ignoring the brightness completely by setting all the “B” values to 50% and leaving the “H” and “S” unchanged, we can better see the color differences without being overwhelmed by the illumination changes.
Colors with especially low saturation are lost, but this shows that the only color shift is subtle: Pure blue shifts in saturation, green shifts a little in hue value, and cyan does both. This is not something that will be noticed, so in terms of color quality of the flash I would say there is no real difference.
Note that this uses the correct white balance adjustment for each flash, individually. What about using both at the same time, so the light sources are mixed?
This adds a yellowing of the mid to dark neutrals, and a hue shift in the magentas. The lightest neutrals don’t have a visible difference, and I think that would be where it would be more bothersome. I expect the overall effect is lost in the normal color casts you pick up from other objects in the room and whatever you are bouncing off or diffusing through.
Both flashes have non-uniform light beams that I believe are the dominant source of differences. In particular, the apparent difference in exposure and the visibly different texturing of the wall is due, I conjecture, to the differences in illumination angles. A difference in diffusion, or apparent size of the light source, can affect the modeling of textures especially when the flash is near the lens. This deserves further research.
Although I saw the in-camera histogram was quite different with one seemingly offering more dynamic range in the exposure, and the reflected light curves are also rather different in some interesting ways, they both work to reproduce colors well. (Assuming a silver-halide technology ANSI standard IT8 calibrating target contains pigments that respond to different spectra of light in a similar manner to real-world scenes, though it was produced with the intention of mimicking the response of various prints and films. But pigments and dies used for making prints are themselves expected to look right under different lighting conditions.)