Level of Detail Study
Level of detail (LOD), or relative sizing, is a compositional concept that describes the size of an object, brush stroke, or another visual element relative to the overall canvas size.
LOD is something I’ve stumbled upon over the past 10 or so years, while grappling with these sorts of questions:
Why do some landscape photos have the power to stop me in my tracks and force me to get lost into it? And why don’t my photos do that?
Why do some of Sharpie sketches on sticky notes and scratch paper project a ton of energy, but then look stale and sterile after enlarging to 8x10”?
Why certain artworks encourage the viewer to inspect close up … and others don’t?
Spectrum Analysis and Level of Detail (LOD)
A good place to start with the LOD concept is spectrum analysis. There is mathematical method called Fourier analysis that is built on the idea that every curve, no matter how complicated, is just a combination of sine waves of varying frequencies. The entire collection of sinusoids for a particular waveform represents the waveform’s spectrum. Reference Wikipedia entry for more details.
In images, a sinusoid manifests as a gradient from white to black, and then from black back to white. If this transition from white to black to white takes the entire width of the image to occur, that is a slow sinusoid. If the transition from white to black to white occurs in just fraction of the image’s width, that is a fast sinusoid.
Figure 1 demonstrates the concept of using slow and fast sinusoids to quantify the level of detail on images. The topmost gradient is very slow, and can only cover 1/10th of the white-black-white gradient. The second image is faster, and can cover the white-black-white gradient within the entire image width. And the third, fourth, and fifth gradients are much faster than the first 2 gradients, and can complete white-black-white gradient 10, 100, and 1000 times over the width of the image.
Figure 1
Five (5) sinusoids of increasing frequencies. In images, 1 cycle of a sinusoid manifests as a white-black-white gradient. The top most sinusoid is slow (i.e. frequency), and is only able to complete 10% of the white-black-white cycle over the width of the image. The second sinusoid completes exactly 1 white-black-white cycle over the width of the image. The lower 3 sinusoids are much faster (i.e. high frequency) and are able to cover 10, 100, and 1000 white-black-white cycles over the width of the image.
Any object within an image (e.g. a shape, mountain, noise and film grain) can be represented as a combination of these white-black-white gradients. The slow gradients represent big objects in the image, and the fast gradients are the fine details and noise.
We will use a photo of Stirling Falls in New Zealand’s Milford Sound (Figure 2) to demonstrate how these abstract gradients apply to the structure of an image. We can see the green-white-green of the photo creates a very prominent 3-column composition that roughly resembles the second gradient shown in Figure 1. This shows that large compositional elements are constructed from low frequency (i.e. slow) sinusoids. Then when we examine the wispy parts of the waterfall or the greenery around the waterfall, we see many fine details that carry the characteristics of the higher frequency (i.e. fast) sinusoids. This shows that fine details and textures are represented by a combination of the faster gradients.
Although this has been an extremely abstract exercise, representing an image’s level of detail as percentage of the overall canvas size provides many benefits in designing a generative algorithm. The rest of this essay examines 3 ratios that provide insight into composition, sizing objects, and selecting stroke widths for a piece of generative art.
Figure 2
Stirling Falls from New Zealand’s Milford Sound
Subject Size to Canvas Ratio
The first LOD ratio we will examine is the subject size to canvas ratio (i.e. length/width of an image’s element divided by the length/width of the overall canvas). The idea here is to understand how to select a size for an image element to achieve different goals (e.g. primary subject, secondary and tertiary subjects, fine detail, texture).
Figure 3 demonstrates the perceptual effect of varying the size of an object relative to the canvas size. My takeaways:
100% large compositional elements or backgrounds
50% primary subjects
5-10% details and secondary subjects
1% or smaller texture, fine detail, and noise
Figure 3
This scribble’s diameter is set relative to the canvas size. From left to right, top to bottom, the scribble’s diameter varies across 100, 50, 10, 5, 1, and 0.5% of the canvas width.
Stroke Width to Subject Size Ratio
The second LOD ratio we will examine is the stroke width to subject size ratio (i.e. stroke width of an object divided by the length/width of that object). The idea here is to understand how stroke width affects the perceptual quality of an element in the image.
Figure 4 demonstrates the perceptual effect of varying stroke width relative to an object’s size. My takeaways:
5-10% looks like using marker on sticky note; a very high energy look
0.5-1% looks like using pen or pencil on a letter-sized sheet; feels a bit more serious
0.1% and smaller looks like partially erased pencil; feels wispy and hair-like and nearly unnoticeable
Figure 4
This scribble’s stroke width is set relative to the diameter of the scribble. From left to right, the scribble’s stroke width varies across 10, 5, 1, 0.5, 0.1, 0.05% of the scribble’s diameter.
Stroke Width to Canvas Size Ratio
The third LOD ratio we will examine is the stroke width to canvas size ratio (i.e. stroke width of an object divided by the length/width of the canvas). The idea here is to understand how stroke width affects the perceptual quality of lines in the image.
Figure 5 demonstrates the perceptual effect of varying stroke width relative to the canvas size. My takeaways:
10% looks like marker on sticky note and graffiti-esque
1% looks like pen or pencil on a letter-sized sheet
0.1% looks like pencil sketches on a letter-sized sheet
0.01% looks like partially erased pencil lines, and is nearly unnoticeable
Figure 5
This scribble’s stroke width is set relative to the canvas size. From left to right, the scribble’s stroke width varies across 10, 1, 0.1, 0.01% of the canvas width.
Print Size Considerations
In the 3 previous sections covering the different LOD ratios, the takeaways have been more about the quality of each LOD rather than hard recommendations, because print size must be taken into consideration.
To me, the final print size for a piece of work is one of the most important creative decisions for an artist. Imagine the difference in experience when you see a broad landscape on a profile pic versus a postcard versus a 4x6 feet print … one is indistinguishable, one is underwhelming, and one is immersive. This principle applies to all types of images.
Broadly speaking, I think in terms of the following size buckets:
X-large (furniture size) – 30x60”, 40x60”, etc.
Large (poster size) – 16x20”, 18x24”, 24x26”, 30x30”, etc.
Medium (A3/A4 size) – 8x10”, 11x14”, 11x17”, etc.
Small (handheld size) – 3x5”, 4x6”, 5x7”, etc.
X-small (card and stamp size) – 1/2x1/2”, 1x1”, 2.5x3.5”, etc.
Big prints (i.e. large and extra-large) tend to be enjoyed in 2 stages: distance viewing and inspection. First, the viewer looks at the work from a distance where they can see the entire image without moving their head, and let’s their eyes wander around the entire piece. But very quickly, if the image contains fine detail, the viewer will want to walk closer to inspect the fine details and texture of the image. For this reason, it is necessary to combine large LOD to build up the macro-composition of the image, and very fine LOD to add small details and textures for inspection. The image should almost have a fractal-esque quality that makes it feels like there is always more and more to see.
Medium prints (i.e. approximately A3 and A4) occupy a sweet spot in viewing. This size range fits within our vision’s entire field of view without being too small. That means the viewer can appreciate fine details at the same time as the overall big picture composition. There is a lot of flexibility for designing the LOD for works at this scale.
Small prints (i.e. small and extra-small) are best enjoyed for the layout and primary subjects. Card-, stamp-, and profile-picture-sized images occupy a very small portion of our vision’s field of view. This means, we primarily notice big shapes at this scale, and fine details and texture become a distraction if it isn’t carefully managed.
Although this section has emphasized physical prints, these principles apply equally to digital displays. The one caveat to consider is: digital display (as of 2022) lack the resolution of paper prints, especially at larger sizes.
Lessons for Composition / TL;DR
LOD is the relative size of a visual element with respect the other image elements or the overall canvas. This essay examined 3 specific LOD ratios:
Subject size to canvas size
Stroke width to subject size
Stroke width to canvas size
The final print size of an image drives how the viewer will experience the artwork, and is one of the most important creative decisions an artist must make.
LOD, in conjunction with final print size, is a core compositional pillars for designing images.
There is one size fits all best LOD or object size or stroke width. The key lies in matching the LOD with the intention of the artwork.