Drop Shadow Mode
drop-shadow-mode
The drop-shadow-mode parameter chooses how a cast shadow works out the subject’s shape and the camera angle. It takes manual or ai and defaults to manual.
Under manual, the shadow is built from the values you give it — chiefly drop-shadow-pose. Under ai, a depth model reads the photograph and estimates the subject’s real three-dimensional shape and how it was photographed, which produces a shadow that bends around the subject’s form rather than projecting a flat silhouette.
When to Use ai
ai earns its cost on subjects with real depth — a bottle, a shoe, a bag — where a flat projection looks pasted on. On flat artwork, a logo or a card it makes little difference and manual is the better choice.
In testing the model’s pose estimate was right on 24 of 30 photographs. Flat items shot head-on, such as a phone or a framed print, are the usual misses; set drop-shadow-pose by hand when that happens.
Cost and Limits
ai adds roughly 30 ms per render and runs on a dedicated AI worker — one per server by default — so a burst of requests may queue behind each other.
If the model is unavailable the render falls back to the standard cast shadow rather than failing. X-Imgix-Params still reports drop-shadow-mode=ai, so a fallback is not visible in the response.
On animations ai falls back to the standard cast shadow for every frame.
One model run serves both an ai shadow and an ai reflection, so combining them does not double the cost.
drop-shadow-mode is read only by drop-shadow-type=cast. An unknown word falls back to manual and is left out of the X-Imgix-Params response header.