[Experiment] Recreating an Amateur Selfie Feel with Prompts

[Experiment] Recreating an Amateur Selfie Feel with Prompts

Theme

In contrast to professional photography, this experiment tests whether a “amateur feel” — like taking a selfie with a smartphone — can be produced through prompts. Keywords like mirror selfie, amateur snapshot, and messy composition are added incrementally, observing how composition, background, and image quality change.

Conclusions First

  • mirror selfie dramatically changes composition and shooting style. Compared to control, a mirror-view composition with a smartphone-holding pose appeared in all 3 seeds.
  • Adding amateur snapshot, messy composition increases background clutter. Clothing and small items scattered around the room reflected in the mirror, producing a lived-in image. The tendency is for it to reflect as “room clutter” rather than “compositional disorder.”
  • The effect of low quality photo, motion blur is limited. Background clutter increased somewhat, but no notable sharpness reduction or motion blur was observed across the whole image.
  • extremely close smartphone selfie achieves face close-up and partial face-hiding by the device. A close-up composition with the smartphone covering part of the face appeared in 3 out of 3 images.
  • lying on bed taking smartphone selfie stably reproduces a lying-on-bed composition. A slightly overhead angle with the smartphone-holding hand in frame — an everyday composition — was obtained.
  • Quality-degradation keywords (film grain, overexposed, out of focus included) have limited effect in z-image-turbo. Only bokeh foreground was partially reflected as background lighting bokeh.

Experiment Design

ItemValue
Modelz-image-turbo (6B, realism-focused distillation)
Steps8
CFG1.0
Size1024x1024
Seed42 / 123 / 789 (common to all conditions)
Per condition3 images (3 seeds)
Total9 conditions × 3 seeds = 27 images

Condition List

ConditionOverviewAdded Keywords
ctrlProfessional shootingprofessional lighting
ABasic mirror selfiemirror selfie, holding smartphone, casual home wear
BAdded amateur feelA + amateur snapshot, messy composition
CAdded low quality feelB + low quality photo, motion blur
DExtreme close selfieextremely close smartphone selfie, face partially hidden by smartphone, messy framing
ELying on bedlying on bed taking smartphone selfie, part of face outside frame, amateur snapshot
FFilm grainfilm grain, noise
GOverexposedoverexposed, washed out colors
HOut of focusout of focus, bokeh foreground

ctrl: Professional Shooting (Baseline)

ctrl: Professional Shooting
1girl, 32yo japanese actress, portrait photo, standing in bedroom, professional lighting
seed 42seed 123seed 789
ctrl seed42ctrl seed123ctrl seed789

Observations: Standing portrait in a bedroom under tidy lighting. Background is organized, with bed and pillows arranged. Subject faces camera directly with a stable pose. Used as the reference for subsequent comparisons.

A: Basic Mirror Selfie

A: Basic Mirror Selfie
1girl, 32yo japanese actress, mirror selfie in bedroom, holding smartphone, casual home wear
seed 42seed 123seed 789
A seed42A seed123A seed789

Observations: All 3 images reproduced a mirror selfie composition with smartphone in hand. Clothing changed to casual home wear (shirt + shorts or jeans), different from the control’s one-piece. Background shows bed and furniture reflected in the mirror, with gaze toward the smartphone screen direction. Compared to control, the pose shifted to a more relaxed posture.

B: Added Amateur Feel

B: Added Amateur Feel
1girl, 32yo japanese actress, mirror selfie in bedroom, holding smartphone, casual home wear, amateur snapshot, messy composition
seed 42seed 123seed 789
B seed42B seed123B seed789

Observations: The mirror selfie composition from Condition A is maintained. The notable change is the background — clothing and small items scattered in the room reflected in the mirror was confirmed in 3 out of 3 images. While Condition A’s background was relatively tidy, adding messy composition resulted in a cluttered, lived-in interior. The subject’s clothing also became more casually impresssive (oversized shirts, etc.).

C: Added Low Quality Feel

C: Added Low Quality Feel
1girl, 32yo japanese actress, mirror selfie in bedroom, holding smartphone, casual home wear, amateur snapshot, messy composition, low quality photo, motion blur
seed 42seed 123seed 789
C seed42C seed123C seed789

Observations: Compared to Condition B, background clutter increased further. More clothing and miscellaneous items scattered in the room, with stronger lived-in feel. On the other hand, no notable sharpness reduction or motion blur from low quality photo or motion blur was observed in any of the 3 images. Lighting became slightly darker, with some images taking on a color tone as if shot under indoor lighting only. Clothing became even more casual than Condition B (T-shirt + distressed jeans, etc.).

D: Extreme Close Selfie

D: Extreme Close Selfie
1girl, 32yo japanese actress, extremely close smartphone selfie at home, casual home wear, face partially hidden by smartphone, messy framing
seed 42seed 123seed 789
D seed42D seed123D seed789

Observations: All 3 images resulted in a close-up face composition with the smartphone covering part of the face (around eye to cheek area). extremely close made the distance to the subject extremely short, with only shoulders upward fitting in the frame. The face partially hidden by smartphone instruction stably reproduced a composition with the device hiding part of the face. Background is barely visible, with the subject’s face, hand, and device occupying most of the frame. This resulted in a completely different composition from A–C (mirror selfie style).

E: Lying on Bed Selfie

E: Lying on Bed Selfie
1girl, 32yo japanese actress, lying on bed taking smartphone selfie, casual home wear, part of face outside frame, amateur snapshot
seed 42seed 123seed 789
E seed42E seed123E seed789

Observations: All 3 images reproduced a lying-on-bed composition with smartphone in hand. From a slightly overhead or sideways viewpoint, pillows, duvet, and sheets occupy most of the frame. The effect of part of face outside frame may have caused some images where the top of the head or forehead is cut off at the frame edge. The lying posture and bedding depiction expressed an everyday scene just before or after sleep.

F: Film Grain + Noise

low quality photo, motion blur didn’t work, so film grain and image noise are tried for quality degradation.

F: Film Grain + Noise
1girl, 32yo japanese actress, mirror selfie in bedroom, holding smartphone, casual home wear, film grain, noise
seed 42seed 123seed 789
F seed42F seed123F seed789

Observations: In seed 42, the mirror selfie composition was reproduced with a gray T-shirt and beige pants as casual home wear. Very slight graininess is visible across the image, though it’s hard to call it clear film grain. The color tone has become somewhat muted with reduced saturation, which could be the effect of film grain. However, the difference from Condition A is limited, and the noise effect was not prominent across all 3 images.

G: Overexposed + Washed Out Colors

Testing for overexposure and faded color feel.

G: Overexposed + Washed Out Colors
1girl, 32yo japanese actress, mirror selfie in bedroom, holding smartphone, casual home wear, overexposed, washed out colors
seed 42seed 123seed 789
G seed42G seed123G seed789

Observations: In seed 42, the mirror selfie composition was reproduced. Overall slightly brighter tone, with some blown-out whites visible in walls and background. Clothing changed to denim shorts, with higher casual degree than Condition A. However, the dramatic blown-out whites from overexposed or clearly faded colors from washed out colors were limited across all 3 images, staying within the range of normal indoor lighting.

H: Out of Focus + Bokeh Foreground

Testing for blur effect.

H: Out of Focus + Bokeh Foreground
1girl, 32yo japanese actress, mirror selfie in bedroom, holding smartphone, casual home wear, out of focus, bokeh foreground
seed 42seed 123seed 789
H seed42H seed123H seed789

Observations: In seed 42, the mirror selfie composition was reproduced. Striped shirt with denim shorts outfit. Round lighting bokeh is depicted in the background, with the bokeh foreground effect partially reflected. However, the subject itself does not have blur applied — the out of focus effect was not confirmed. The bokeh tends to be applied to background lighting rather than the foreground.

Summary Comparison

ConditionComposition changeBackground clutterSmartphoneImage quality change
ctrlFront portraitNone (tidy)NoneNone
AMirror selfieMinorHeld in one handNone
BMirror selfieProminent (clothing scattered)Held in one handNone
CMirror selfieEven more prominentHeld in one handSlightly darker lighting
DFace close-upAlmost invisiblePartially covers faceNone
ELying on bedNone (bedding-centered)Held in one handNone
FMirror selfieMinorHeld in one handVery slight graininess
GMirror selfieMinorHeld in one handSlightly brighter
HMirror selfieMinorHeld in one handBackground bokeh present

The tendency for messy composition to act in the direction of “cluttering the background” rather than “disrupting the composition” was confirmed. For changing the composition itself, keywords that directly specify pose and framing like extremely close, lying on bed, and face partially hidden were more effective.

low quality photo and motion blur were not clearly reflected as quality degradation or blur in 18 generated images. film grain, noise, overexposed, washed out colors, and out of focus were similarly of limited effect, with no dramatic image quality changes confirmed. Only bokeh foreground was partially reflected as background lighting bokeh. Quality-degradation keywords in general have limited effect in z-image-turbo’s 8-step, CFG 1.0 environment.

Lab Director Comment

So, the biggest discovery this time is that messy composition affects the room’s clutter degree rather than composition. To disrupt the composition, it seems more effective to directly specify pose and framing with things like extremely close or face partially hidden. D’s face-hiding selfie seems useful, and E’s lying-on-bed is also stable and good for conveying everyday feel. The fact that quality-degradation keywords generally don’t work is something I want to remember as a characteristic of this model.