How AI interprets · 3 min read
How AI interprets light.
On the single-source raked-light specification — why it matters, and why models default to "ambient" lighting that has no source at all.
Image models default to soft ambient lighting with no identifiable source — the visual equivalent of "everywhere and nowhere." This produces images that look pleasant but cannot be read as photographs, paintings, or specimen plates. Specifying a single light source, its angle, and its temperature is the cheapest brief intervention that improves output quality the most.
Light is the variable that does the most work in a fine art image and the variable that AI most consistently fudges.
What the model defaults to
Soft, sourceless, ambient — like a photograph taken in an overcast room with no windows. Pleasant, forgettable, untraceable. The shadow it casts has no direction; the highlights it makes have no logic.
What a brief specifies
- Source direction (canvas-left, upper-right, etc.)
- Source temperature (sodium amber ~2200K, daylight ~5500K, etc.)
- Source intensity relative to ambient (raked = high contrast, diffuse = low contrast)
- Shadow handling — is the shadow part of the composition or a side effect?
- For multi-source plates: which source dominates, which subordinates
Why this matters
A specified light source produces an image that reads as a real artifact — a painting under museum light, a photograph at a specific time of day, a plate from a specific tradition. Unsourced light reads as a render. The buyer notices, even if they cannot articulate what is wrong.