← Back to App
← All tag comparisons

Lens and tilt

What does each of the tags that bend or tilt the frame actually touch?

The previous three comparisons moved the camera. This one changes how the camera distorts what it sees.

They differ in force. Two bend geometry hard, one relocates the viewpoint, and the last bends nothing at all — it changes the relationship instead.

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

dutch angle
The horizon tilts. Buildings lean and the frame turns unstable — the grammar of thrillers and action.
fisheye
Barrel distortion curves the edges. Straight lines bow and the centre swells. The strongest effect of the four.
drone view
No aerial view came out. The camera looks down only slightly from about head height; it is almost the same picture as bird's-eye view in the camera-height set, and no higher than from above.
pov
Bends no geometry. Instead the camera moves a little closer, the figure fills more of the frame and the gaze meets the viewer. Set beside the straight-on cut in the subject-facing comparison (same seed), though, that is the whole difference, and the two look quite alike.

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

dutch angle and fisheye fight each other in one frame. pov sits on a different axis and layers with the rest.

How far this result reaches

On danbooru, pov carries an implication that the viewer is part of the scene. Combined with other tags it can drift somewhere you did not intend — check the output.

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.

Ready to build your prompt?

Click from 1,700+ tags to assemble your AI image prompt.

Start Prompt Builder →