[Experiment] Bangs Prompt Test | Comparing Generation Accuracy Across 10 Styles

Conclusions First

blunt bangs had the highest control precision. All 3 images showed thicker, denser, more horizontal bangs than the control — a clear difference that makes it the top recommended bangs prompt.
baby bangs is also highly stable. All 3 images showed shorter bangs, making it a strong alternative to blunt bangs.
no bangs increases forehead exposure but doesn’t fully remove bangs. Some seeds retain a light fringe.
curtain bangs / side-swept bangs / see-through bangs show little visual difference from the control. The default bangs in the model already resemble these styles, making the effect hard to confirm.
hair over one eye covered one eye with hair in 2 out of 3 images — moderate control. Reduces face exposure, which is disadvantageous for expression rendering.
choppy bangs causes the entire hair to become shorter as a side effect. All 3 images showed hair shortening to about a short bob despite specifying medium hair.

Experiment Design

Parameter	Value
Model	z-image-turbo (6B, photorealistic distillation)
Steps	8
CFG	1.0
Resolution	1024×1024
Conditions	11 (including control)
Seeds	3 (fixed across all conditions)
Total images	33
Changed variable	Bangs prompt only (1 variable)

Base Prompt

1girl, 32yo japanese actress, medium hair, {BANGS_STYLE}, standing, facing camera, upper body, simple white background, soft studio lighting

{BANGS_STYLE} is replaced with each condition’s prompt. Control (A00) has no bangs specification. Hair length is fixed at medium hair to clearly observe bangs changes.

Evaluation Criteria

Shape accuracy — how well the output matches the specified bangs shape (○: clear match / △: partial match / ×: mismatch)
Stability — number of matches across 3 seeds (3/3, 2/3, 1/3, 0/3)
Face visibility — how bangs affect the appearance of the face (eye/forehead exposure, etc.)
Side effects — unintended effects on elements beyond bangs (overall hairstyle, outfit, etc.)
Recommendation — overall rating (★★★ / ★★ / ★ / −)

Results

A00: Control (No Bangs Specification)

A00: control

1girl, 32yo japanese actress, medium hair, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: All 3 images show bangs at a natural length reaching the eyebrows, with a center part or slightly swept style. Forehead partially visible. seed 3 shows an umbrella and non-white background. This serves as the baseline for default bangs style to compare against subsequent conditions.

A01: blunt bangs

1girl, 32yo japanese actress, medium hair, blunt bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: All 3 images show bangs cut horizontally at the eyebrow level (blunt cut). Compared to the control, bangs are denser and forehead exposure is reduced. seed 3 has bookshelves appearing in the background as a side effect, but bangs shape is clearly controlled.

Note: blunt bangs reliably reproduces blunt bangs across all 3 images, making it directly usable for the scene-kid (jirai-kei) aesthetic.

A02: side-swept bangs

1girl, 32yo japanese actress, medium hair, side-swept bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: All 3 images show bangs swept to one side. However, similar bangs appear in the control (A00) as well, making it hard to determine if this is a prompt-specific effect. seed 3 shows an umbrella. No clear difference from the control is confirmed.

A03: curtain bangs

1girl, 32yo japanese actress, medium hair, curtain bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: 2 out of 3 show bangs parting left and right from center. seed 2 shows a composition as if shot through a curtain. seed 3 shows a curtain appearing in the background as a side effect. The part itself is visible but the difference from control is small. “curtain” can be interpreted as background curtains.

A04: see-through bangs

1girl, 32yo japanese actress, medium hair, see-through bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: 2 out of 3 show thin bangs with skin visible through them. The remaining 1 is similar density to the control. seed 3 shows both arms raised as a side effect. Difference from control is small, but a slightly lower bang density is confirmed.

A05: no bangs

1girl, 32yo japanese actress, medium hair, no bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: 2 out of 3 show more forehead exposure than the control, with bangs swept sideways revealing the upper forehead. However, seed 3 retains normal bangs — complete “no bangs” is not achieved. Works in the direction of increasing forehead exposure, but cannot fully eliminate bangs.

A06: baby bangs

1girl, 32yo japanese actress, medium hair, baby bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: All 3 images show bangs at a short position above the eyebrows. Clearly shorter than the control, with the upper half of the forehead exposed. seed 3 shows illustrations appearing in the background as a side effect. The shortening effect is stable, though it leans more toward naturally short bangs rather than the precise horizontal line of “micro bangs.”

A07: asymmetrical bangs

1girl, 32yo japanese actress, medium hair, asymmetrical bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: No clear difference from control. All 3 images show naturally flowing bangs with no noticeable left-right asymmetry. seed 3 shows illustrations in the background. The asymmetrical bangs prompt has no confirmed effect on bangs — the result is the same as the control.

A08: wispy bangs

1girl, 32yo japanese actress, medium hair, wispy bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: No clear difference from control. All 3 images show naturally flowing bangs. The airy, gapped look cannot be confirmed — bang density is similar to control. seed 3 shows a dragon illustration appearing in the background. Effect of wispy bangs prompt is weak.

A09: choppy bangs

1girl, 32yo japanese actress, medium hair, choppy bangs, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: All 3 images show the overall hair becoming shorter than specified by medium hair, ending up around short-bob length. The bangs themselves are short and slightly uneven, but not the rough-cut look you’d call “choppy.” The side effect of “choppy” affecting overall hair length is prominent. seed 3 shows a floral umbrella in the background.

A10: hair over one eye

1girl, 32yo japanese actress, medium hair, hair over one eye, standing, facing camera, upper body, simple white background, soft studio lighting

seed 1	seed 2	seed 3

Observation: 2 out of 3 show bangs covering one eye (seed 1, seed 3). seed 2 has bangs hanging between both eyes but not fully covering one eye. The face exposure area decreases when one eye is hidden by hair. seed 3 shows a knitting prop in the background. Hair over one eye works with moderate precision but has some instability.

Summary

All Conditions Summary

Condition	Prompt	Shape Accuracy	Stability (3 images)	Face Visibility	Side Effects	Recommendation
A00	(none)	—	—	Forehead partially visible	—	—
A01	blunt bangs	○	3/3	Forehead hidden by bangs	Small (background change)	★★★
A02	side-swept bangs	×	0/3	Same as control	None	−
A03	curtain bangs	△	2/3	Center of forehead visible	Medium (curtain in background)	★
A04	see-through bangs	△	2/3	Forehead somewhat visible	None	★★
A05	no bangs	△	2/3	Increased forehead exposure	None	★★
A06	baby bangs	○	3/3	Upper forehead exposed	Small (background change)	★★★
A07	asymmetrical bangs	×	0/3	Same as control	None	−
A08	wispy bangs	×	0/3	Same as control	None	−
A09	choppy bangs	△	3/3	Forehead slightly visible	Large (entire hair becomes short)	−
A10	hair over one eye	○	2/3	One eye hidden by hair	Small (background change)	★★

Recommended Prompts

Goal	Recommended Prompt	Notes
Want blunt/straight bangs	`blunt bangs`	Highest precision and stability. First choice.
Want shorter bangs	`baby bangs`	Creates bangs above the eyebrows. Upper forehead exposed.
Want to cover one eye	`hair over one eye`	2/3 precision. Affects composition.
Want thinner bangs	`see-through bangs`	Small effect but tends to lower bang density vs. control.
Want no bangs	`no bangs`	Complete removal is difficult, but forehead exposure increases.

Prompts to avoid:

side-swept bangs — no confirmed difference from control
asymmetrical bangs — no confirmed difference from control
wispy bangs — no confirmed difference from control
choppy bangs — large side effect of shortening entire hair

Note: For bangs, blunt bangs and baby bangs are the reliable choices. With 4 keywords confirmed ineffective, knowing these 2 is sufficient.
PR RunPod クラウドGPUでAI画像生成 RunPodを始める →