Camera Angles & Framing
Control visual impact with angle (low/high/eye level) and framing (close-up/full shot) tags.
1. Angle vs framing
Two independent axes describe any shot in AI prompting. Angle is where the camera sits relative to the subject — from above / from below / straight-on. Framing is how much of the subject is in the frame — close-up, medium shot, full body. Specify both for predictable compositions. Skipping the angle tag lets the model default to eye-level medium shots, which is usually safe but also the most generic choice.
2. Vertical angles
low angle / from belowCamera below the subject, looking up. Makes the subject feel powerful, heroic, or intimidating. Good for hero shots, mecha, and monuments. Caveat: elongates legs in portrait photos.
high angle / from aboveCamera above looking down. Makes subjects feel small, vulnerable, or observed. Works well for chibi and cute compositions.
straight-on / eye levelNeutral, conversational. Default for portraits. The two mean the same thing, but straight-on is the danbooru spelling and the safer pick on booru-trained checkpoints — eye level drew no response on the Illustrious model we tested, while generators that read plain English accept either. Explicitly tag it if your subject is kneeling or floating — otherwise the model may drop the camera automatically to follow them.
bird's eye view / top-down viewDirectly above. Classic for flat-lay food photography, maps, and game UI references. Extreme version: overhead shot.
worm's eye viewDirectly below, looking up. Dramatic for skyscrapers, forest canopies, and fight scenes.
from aboveFrom above
from belowFrom below
bird's-eye viewBird's-eye view
worm's-eye viewWorm's-eye viewShared 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>
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
This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.
3. Tilted and dynamic angles
dutch angle / canted angleCamera tilted sideways. Feels unstable, chaotic, anxious. Signature of action scenes, horror, and noir. Pair with motion blur.
over the shoulderCamera behind one character looking at another. Common in dialogue scenes and video games. Combine with a foreground subject: over the shoulder, blurry foreground figure.
pov / first person viewFrom the subject's eyes. Useful for games, UI mockups, and intimate scenes.
straight-onStraight-on
from sideFrom side
from behindFrom behind
profileProfileShared 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>
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
This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.
4. Framing — how much to show
- •
extreme close-up— eyes, lips, hands only. For emotional emphasis. - •
portrait/close-up— head and shoulders. Default for portraits. - •
upper body/medium shot— chest to waist up. Good for dialogue framing. - •
cowboy shot— mid-thigh up. Named for Western films; shows holsters/hands. - •
full body— head to toe. For character sheets, outfit reveals, poses. - •
wide shot/long shot— subject and environment visible. - •
extreme wide shot/establishing shot— subject is tiny, environment dominates.
Two vocabularies are mixed in that list. portrait, upper body, cowboy shot and full body come from danbooru captions; the film terms beside them are ordinary English. Generators that read natural language — NovelAI, Midjourney — handle both. Booru-trained Stable Diffusion checkpoints lean on the first set: on the Illustrious checkpoint we tested, the four danbooru terms produced a clean four-step ladder (see the grid) while medium shot and long shot barely moved the frame. If a framing word seems to do nothing, try its danbooru equivalent before assuming you wrote it wrong.
📐 Angle / Shot — Tag examples
close-up — Close-up
portraitPortrait
upper bodyUpper body
cowboy shotCowboy shot
full bodyFull bodyShared prompt
masterpiece, best quality, amazing quality, very awa, 1girl, solo, mature female, adult woman, short black hair, brown eyes, white blouse, city street, daytime, clear sky, <tag slot>
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
This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.
5. Lens and depth cues
Framing combines with lens vocabulary to control how the image feels:
- •
wide angle lens,fisheye— exaggerated perspective, curved edges - •
85mm lens,portrait lens— compressed, flattering faces - •
telephoto,400mm— subject isolated, background compressed - •
macro— extreme close-up with nature photography feel - •
shallow depth of field,bokeh,f/1.8— blurred background - •
deep depth of field,everything in focus— landscape clarity
dutch angleDutch angle
fisheyeFisheye
drone viewDrone view
povPOVShared 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>
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
This comparison was measured on a single Stable Diffusion family checkpoint. The same tags can behave differently on NovelAI, Midjourney or another checkpoint — read it as a reference point, not a verdict.
6. Common pairings
Heroic introduction
low angle, full body, dramatic lighting, wide angle lens, epic scenery
Intimate portrait
straight-on, portrait, 85mm lens, f/1.8, soft lighting
Action chase
dutch angle, medium shot, motion blur, wide angle lens, running
Flat-lay food
bird's eye view, top-down, food photography, natural light, square framing
7. Mistakes to avoid
- Conflicting angle tags. Don't combine
low anglewithbird's eye view— the model picks one, often the wrong one. - Framing without angle.
portraitalone leaves the angle to interpretation. Specifystraight-on, portraitfor a predictable result. - Full body + 85mm lens. 85mm is a portrait focal length; for full body prefer 35mm or wide angle.
- Ignoring aspect ratio. Angle and framing only make sense within an aspect ratio. Portrait 2:3 rewards vertical angles; 16:9 rewards wide environmental framing.
Related: composition tags · Angle tags · Composition tags