Key Findings
crawling on bedis the most stable pose description — reproduced the all-fours + butt-out composition in 2 out of 3 seedsass uprisks generating two people (occurred in 1/3)from behindandrear vieware equivalent for angle. Addinglow angleincreases impactslightly aboveinduces a prone position — avoid it for butt-focused compositions- Lingerie type affects pose stability — details in Experiment 3
What You’ll Learn
- The most stable pose description for rear all-fours compositions
- Output differences between angle keywords (from behind / rear view / low angle / slightly above)
- How lingerie type affects outfit reproduction and pose stability
Test Conditions
| Parameter | Value |
|---|---|
| Model | z-image-turbo (6B, photorealistic distilled model) |
| Steps | 8 |
| Sampler | euler |
| Scheduler | ddim_uniform |
| CFG | 1.0 |
| Image Size | 1024×1024 |
| Seeds | 42, 123, 789 (fixed) |
Experiment 1: Pose Description Comparison
Condition A: on all fours (3 tokens)

Observations
Only 1 out of 3 (seed 42) produced the all-fours + looking back composition. The other 2 resulted in near-prone positions. on all fours alone is unreliable when combined with rear angles.
Condition B: on all fours, arched back (5 tokens)

Observations
Adding arched back improves the butt-out feel. However, actual all-fours stability is similar to Condition A (1/3). The back arch was visible in all 3 outputs.
Condition C: on all fours, ass up (5 tokens)

Observations
Seed 42 generated two people despite the 1girl specification. ass up appears to destabilize composition interpretation. The remaining 2 produced kneeling rear views — butts are emphasized but not in all-fours position.
Condition D: on hands and knees, presenting (5 tokens)

Observations
All 3 produced kneeling poses instead of all-fours. presenting also caused lace lingerie set to be interpreted as a bodysuit in 2/3 seeds. Be aware this keyword influences outfit interpretation.
Condition E: crawling on bed (3 tokens)

Observations
2 out of 3 produced natural crawling + looking back compositions. crawling on bed has the best compatibility with rear butt-out framing, and the pose feels natural. Token efficiency is also excellent at 3 tokens.
Lab Director: Crawling wins again. Telling the model to “crawl” works better than saying “get on all fours” — the action-based description produces more natural poses. Plus it adds a looking-back motion for free. Love it.
Pose Comparison Table
| Seed | on all fours (A) | arched back (B) | ass up (C) | presenting (D) | crawling (E) |
|---|---|---|---|---|---|
| 42 | ![]() | ![]() | ![]() | ![]() | ![]() |
| 123 | ![]() | ![]() | ![]() | ![]() | ![]() |
| 789 | ![]() | ![]() | ![]() | ![]() | ![]() |
Experiment 2: Angle Description Comparison
Condition A: from behind (2 tokens)

Observations
Standard rear angle. Pose variety is wide — mix of all-fours, kneeling, and prone positions.
Condition B: rear view (2 tokens)

Observations
Nearly identical output to from behind. Same token count (2 tokens). No practical difference observed.
Condition C: from behind, low angle (4 tokens)

Observations
low angle increases butt impact. Seed 123 showed wider leg spread, creating a more dynamic composition. However, outfit was interpreted as bodysuit in 1/3 cases.
Condition D: from behind, slightly above (4 tokens)

Observations
slightly above produced prone positions in all 3 seeds. The overhead angle destabilizes the all-fours pose. Avoid this for butt-out compositions.
Angle Comparison Table
| Seed | from behind (A) | rear view (B) | low angle (C) | slightly above (D) |
|---|---|---|---|---|
| 42 | ![]() | ![]() | ![]() | ![]() |
| 123 | ![]() | ![]() | ![]() | ![]() |
| 789 | ![]() | ![]() | ![]() | ![]() |
Experiment 3: Lingerie Type Comparison
Condition A: lace lingerie set (3 tokens)

Observations
Bra + panties set output consistently. Lace detail reproduction is high. Color varies by seed (navy or white).
Condition B: sheer babydoll nightgown (3 tokens)

Observations
Translucent babydoll is well reproduced. The butt view differs significantly from bra + panties sets — the gown hem drapes over the butt area.
Condition C: black thong and matching bralette (5 tokens)

Observations
The thong specification maximizes butt exposure compared to other conditions. Excellent compatibility with rear compositions — butt lines are clearly defined.
Condition D: white lace bodysuit (3 tokens)

Observations
One-piece bodysuit output is consistent. The silhouette is elegant because it covers the entire butt. Lace transparency provides a good balance of skin visibility.
Condition E: garter belt and stockings with lingerie (6 tokens)

Observations
Adding garter belt and stockings increases visual complexity. The garter line at the thigh-butt boundary creates a natural eye-guiding effect. Token cost is high at 6 but the visual payoff is substantial.
Lab Director: That garter line cutting across the thigh-to-butt transition? It’s like a visual arrow pointing right where you want to look. Worth every token.
Lingerie Comparison Table
| Seed | lace set (A) | babydoll (B) | thong (C) | bodysuit (D) | garter (E) |
|---|---|---|---|---|---|
| 42 | ![]() | ![]() | ![]() | ![]() | ![]() |
| 123 | ![]() | ![]() | ![]() | ![]() | ![]() |
| 789 | ![]() | ![]() | ![]() | ![]() | ![]() |
Summary
Pose Description
| Description | Tokens | Stability | Butt-Out Rate | Notes |
|---|---|---|---|---|
on all fours | 3 | Low | 1/3 | Tends to become prone |
on all fours, arched back | 5 | Low-Med | 1/3 | Back arch is visible |
on all fours, ass up | 5 | Low | 2/3 | Risk of generating 2 people |
on hands and knees, presenting | 5 | Low | 0/3 | Becomes kneeling; changes outfit |
crawling on bed | 3 | Med-High | 2/3 | Best pick. Natural look-back included |
Angle Description
| Description | Tokens | Effect |
|---|---|---|
from behind | 2 | Standard rear. Stable |
rear view | 2 | Equivalent to from behind |
from behind, low angle | 4 | More butt impact. Wider leg spread |
from behind, slightly above | 4 | Induces prone. Not recommended |
Lingerie Type
| Description | Tokens | Butt Visibility | Character |
|---|---|---|---|
lace lingerie set | 3 | Through panties | Stable. Color varies by seed |
sheer babydoll nightgown | 3 | Through gown hem | Translucent. More fabric coverage |
black thong and matching bralette | 5 | Maximum | Best for butt line emphasis |
white lace bodysuit | 3 | One-piece silhouette | Elegant. Good lace transparency |
garter belt and stockings with lingerie | 6 | Garter line accent | Eye-guiding effect. Token-heavy |
Recommended Prompt
- Pose:
crawling on bed(most stable, natural look-back) - Angle:
from behind(2 tokens is enough) - Outfit: Choose by goal (butt emphasis →
black thong, elegance →lace bodysuit)
Lab Director: Crawling wins, from behind is enough, thong seals the deal. Simple but correct. Short prompts hit harder — proven once again.
![[Experiment] All Fours Pose Test — Effects of Pose Keywords and Angle Combinations](/tips/all-fours-pose-test/A_seed42_0_0000_42.webp)
![[Test] All Fours × Cleavage Prompt Comparison — Effects of cleavage, from above, and crawling](/tips/all-fours-cleavage-test/E_seed42_0_0000_42.webp)


![[Verified] Image Generation Prompt Best Practices](/tips/prompt-best-practices/cover_0_0000_4517457392071889496.webp)


