Camera height — from above, from below
How different are the four tags that move the camera up and down?
Vertical angle decides how much power the subject holds. A camera below makes them loom; a camera above makes them small. That much is obvious in words — what is not obvious is how far each tag actually moves the frame.
The four cuts below share a seed, a base prompt and a sampler, and differ by one tag. The two low tags land in nearly the same place, while on the high side bird's-eye view looks down less than from above, despite its name — that is the point of this comparison.
Every cut below shares one seed, one base prompt and one sampler setting, and differs only by the tag marked in amber.
What each cut shows
from above- The camera looks down from above eye level. Shoulders and the crown of the head fill more of the frame and the street falls away below. The downcast gaze puts the subject on the observed side of the exchange.
from below- The camera drops below the waist and looks up. The figure rises against the frame and the buildings tilt back behind them. This is the basic grammar of a hero shot.
bird's-eye view- Despite the name, it looks down only slightly, from lower than
from above. For a higher overhead view,from abovecame closer in this set. worm's-eye view- Lands almost on top of
from below. If you wanted the extreme ground-level look, you need tags that push perspective on top of it.
How these were made
Shared prompt
masterpiece, best quality, amazing quality, very awa, 1girl, solo, mature female, adult woman, short black hair, brown eyes, white blouse, beige trench coat, black trousers, standing, city street, daytime, clear sky, <tag slot>
Tap a tag to carry it over to the builder
Negative prompt
school uniform, pleated skirt, blazer, serafuku, child, loli, teenage, bad quality, worst quality, worst detail, sketch, censor, jpeg artifacts, watermark, signature, text, extra limbs, bad hands, bad anatomy
Model
rinFlanimeIllustrious_v30 · euler_ancestral/normal · steps 28 · CFG 5 · 832×1216 · seed 777001
Takeaway
In practice you have two tools: from above and from below. worm's-eye view landed almost on top of from below, and bird's-eye view actually came out lower than from above, so neither turned out to be a stronger version. Rather than stacking all four, use the two and control the degree with background and perspective tags.
How far this result reaches
On generators that read natural language, bird's-eye view may separate much more clearly. This is one danbooru-trained checkpoint responding.
Measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint. This is a reference point, not a verdict.