[Summary] Skeleton Type × Outfit Compatibility Test | Cross-Comparison, 33 Images

Conclusions First

• No clear body silhouette differences were confirmed from body type keywords (athletic straight body frame / slim wavy body frame / natural body frame, etc.). Comparing same-seed images across C00–C03, there are minor differences in facial features, hair style, and pose, but zero systematic differences in shoulder width, waist, or hip silhouette were confirmed in 3 out of 3 images
• Outfit keyword changes were correctly reflected. C04 produced tailored jacket + pencil skirt; C05 produced chiffon blouse + flare skirt; C06 produced oversized knit + wide-leg pants — each in 3 out of 3 images
• Deep-dive test: Concrete body descriptors (wide hips, broad shoulders, tall, petite, etc.) also produced no body silhouette changes. Zero effect was confirmed across 12 images — identical results to abstract keywords
• Body type differentiation is currently difficult with keywords, but outfit differentiation works well

What You’ll Learn

• Whether skeleton straight/wave/natural type specifications actually change silhouette
• How visible body differences are with the same outfit (white T-shirt + denim)
• Whether concrete body descriptors (wide hips, broad shoulders, petite, etc.) change body silhouette
• Whether specifying outfits “suited” to each skeleton type produces convincing images
• Cross-comparison based on findings from individual tests (straight/wave/natural) — see related articles below

Experiment Design

Base Prompt (Body Type Comparison)

Standardized to white T-shirt + fitted jeans — only the body type keyword changes. White background for clearer silhouette visibility.

Condition List

Body Type Comparison (Same Outfit)

C00: Control (No Body Type Specified)

seed 1	seed 2	seed 3

Observation: All 3 generate white T-shirt + skinny-fit jeans as specified. Background is gray textured wall. All 3 show a standard slim build with no notable shoulder/waist/hip characteristics. Seed 1 has clear sandals, seed 2 has black strap sandals, seed 3 is barefoot.

C01: Straight Type (athletic straight body frame)

seed 1	seed 2	seed 3

Observation: All 3 generate white T-shirt + skinny jeans. Comparing same seeds with C00: minor facial and hair differences (seed 2 has straighter, slightly longer hair; seed 3 has slightly different bangs). However, zero clear differences in shoulder/waist/hip silhouette were confirmed in 3 out of 3. No body type change observed from adding “athletic straight body frame.”

C02: Wave Type (slim wavy body frame, narrow shoulders, wide hips)

seed 1	seed 2	seed 3

Observation: All 3 generate white T-shirt + skinny jeans. Minor facial/hair differences from C00 (seed 2 has wavy long hair). Zero systematic differences in shoulder/waist/hip silhouette were confirmed in 3 out of 3. No shoulder narrowing or hip expansion observed from adding “slim wavy body frame, narrow shoulders, wide hips.”

C03: Natural Type (natural body frame, broad shoulders, angular bone structure)

seed 1	seed 2	seed 3

Observation: All 3 generate white T-shirt + skinny jeans. Minor facial/hair differences from C00 (seed 1 gives a slightly broader shoulder impression, but within the range of pose/angle differences). Zero systematic differences in shoulder/waist/hip silhouette were confirmed in 3 out of 3. No angular bone structure observed from “natural body frame, broad shoulders, angular bone structure.”

Body Type Comparison Table (Same Seed Side-by-Side)

Condition	seed 1	seed 2	seed 3
C00: No specification
C01: Straight
C02: Wave
C03: Natural

Outfits Suited to Each Skeleton Type

Here we test whether combining each skeleton type with “suited” outfits produces convincing images.

C04: Straight × Tailored Jacket

Skeleton straight is said to suit outfits with straight lines. Testing with a tailored jacket.

seed 1	seed 2	seed 3

Observation: All 3 generate tailored jacket + white blouse + pencil skirt as specified. Seed 1 has a gray jacket + matching skirt, seed 2 is navy-gray, seed 3 is light gray. The jacket features a fitted silhouette nipped at the waist, forming a straight line with the knee-length pencil skirt. No clear body silhouette difference from C00.

C05: Wave × Flare Skirt

Skeleton wave is said to suit lightweight, flowing garments. Testing with a flare skirt.

seed 1	seed 2	seed 3

Observation: All 3 generate chiffon blouse + flare skirt as specified. Seed 1 has a geometric-patterned blue chiffon blouse + gray midi flare skirt; seed 2 has a floral navy chiffon blouse + navy midi flare skirt; seed 3 shows a floral chiffon A-line flare silhouette (though framing cuts off at the waist). All 3 show sheer chiffon texture. No clear body silhouette difference from C00.

C06: Natural × Oversized Knit

Skeleton natural is said to suit casual, relaxed clothing. Testing with an oversized knit.

seed 1	seed 2	seed 3

Observation: All 3 generate oversized knit + wide-leg pants as specified. Seed 1 is a beige ribbed knit; seed 2 is gray ribbed knit; seed 3 is beige ribbed knit. All feature a drop-shoulder silhouette covering the hips and wide-leg pants creating a relaxed, straight-line look. Clearly different from C00–C03’s white T-shirt + skinny jeans — outfit specification is working. No clear body silhouette difference from C00.

Suited Outfits Comparison Table (Same Seed Side-by-Side)

Condition	seed 1	seed 2	seed 3
C04: Straight + Jacket
C05: Wave + Flare
C06: Natural + Knit

Deep-Dive Test: Control via Concrete Body Descriptors

Phase 1 showed that abstract body type keywords like “athletic straight body frame” had no effect. Here we test whether directly describing specific physical characteristics can control body silhouette.

Experiment Design (Deep-Dive)

Condition List (Deep-Dive)

D00: wide hips, narrow waist, short torso

seed 1	seed 2	seed 3

Observation: All 3 generate white T-shirt + skinny jeans. Zero hip width expansion or waist emphasis was confirmed in 3 out of 3 images when compared with C00 (control) at the same seed. Even with direct descriptors like “wide hips” and “narrow waist,” no body silhouette change was observed.

D01: broad shoulders, long torso, muscular thighs

seed 1	seed 2	seed 3

Observation: All 3 generate white T-shirt + skinny jeans. Zero shoulder expansion or thigh muscle definition was confirmed in 3 out of 3 at the same seed as C00. Facial features trended toward non-East Asian in all 3 — body type keywords may have interfered with face generation. However, no systematic differences in shoulder width, torso length, or thigh thickness were observed.

D02: tall, long limbs, angular bone structure, flat chest

seed 1	seed 2	seed 3

Observation: All 3 generate white T-shirt + skinny jeans. Zero clear height or limb length differences were confirmed in 3 out of 3 at the same seed as C00. No chest size change from “flat chest” was observed. Even with direct descriptors like “tall” and “long limbs,” no systematic change in subject proportions within the frame was seen.

D03: petite, short, small frame

seed 1	seed 2	seed 3

Observation: All 3 generate white T-shirt + skinny jeans. Compared with C00 at the same seed, some images show the subject appearing slightly smaller in the frame (seeds 2 and 3 have slightly more headroom), but this is within the range of framing/shooting distance differences — zero images systematically express a smaller stature.

Deep-Dive Comparison Table (Same Seed Side-by-Side)

Condition	seed 1	seed 2	seed 3
C00: No specification (reprinted)
D00: wide hips etc.
D01: broad shoulders etc.
D02: tall etc.
D03: petite etc.

Deep-Dive Conclusions

Switching to concrete physical descriptors (wide hips, broad shoulders, tall, petite, etc.) also produced zero confirmed effects on body silhouette in 12 out of 12 images. Results are identical to Phase 1’s abstract body type keywords (body frame series) — for this model, body-related keywords have almost no effect on output body silhouette.

The only observed change: in D01, facial features trended toward non-East Asian in all 3 images. “Broad shoulders” and “muscular thighs” may have acted on racial characteristics rather than body type.

Summary

Results confirmed in this experiment:

1. Outfit keywords were correctly reflected: White T-shirt + jeans for C00–C03; tailored jacket + pencil skirt for C04; chiffon blouse + flare skirt for C05; oversized knit + wide-leg pants for C06 — each output in 3 out of 3 images. Outfit differentiation works well

2. No silhouette differences from body type keywords were confirmed: Zero systematic differences in shoulder/waist/hip between C00 (no body type) and C01–C03 (straight/wave/natural) were confirmed in 12 out of 12 images. Minor differences in facial features and hair are present, but insufficient as body silhouette differentiation

3. Clear outfit differences in suited outfit section (C04–C06): Straight silhouette of tailored jacket, curved silhouette of flare skirt, and relaxed silhouette of oversized knit were each differentiated. However, body type keyword effects were not confirmed even in combination with outfit changes

4. No effect from concrete body descriptors (deep-dive): Changing to specific body part descriptions like wide hips, broad shoulders, tall, petite produced zero confirmed effects on body silhouette in 12 out of 12. Results identical to abstract keywords (body frame series)

5. Body type keywords sometimes interfere with facial features: In D01 (broad shoulders, muscular thighs), facial features trended toward non-East Asian in all 3 images. May have acted on racial characteristics rather than body type

For this model, body-related keywords have almost no effect on output body silhouette. Since outfit changes dramatically alter the silhouette impression, controlling skeleton type expression through outfit selection rather than body type keywords is the practical approach.

Note: Outfit differentiation is working well — tailored jacket, flare skirt, and oversized knit all came out 3/3. Body type is the problem: across 33 images comparing straight/wave/natural, none showed a confirmable silhouette difference. The deep-dive with explicit wide hips and broad shoulders also produced zero effect out of 12. Body type keywords appear ineffective with this model. To suggest a skeleton type impression, outfit selection is the practical route rather than body type keywords.

Related Articles

⚠ 関連記事が見つかりません: /en/tips/skeleton-straight-test

⚠ 関連記事が見つかりません: /en/tips/skeleton-wave-test

⚠ 関連記事が見つかりません: /en/tips/skeleton-natural-test

⚠ 関連記事が見つかりません: /en/tips/cosplay-outfit-comparison