[Experiment] All Fours Pose Test — Effects of Pose Keywords and Angle Combinations

Conclusions

• all fours / on all fours / on hands and knees / crawling all produce essentially equivalent output
• Only crawling tends to produce a slightly more dynamic pose
• Angle specification (from side, from behind, from below) has more impact on pose expression than keyword choice
• Complete reproduction of the all fours pose is unstable (may shift to sitting or kneeling)

What You’ll Learn from This Article

• The effectiveness and differences between all fours-type pose prompts
• Interaction with angle specification (front, side, back, low angle)
• Pose stability and conditions under which it breaks down

Experiment Conditions

Base Prompt

The unified condition is a scenario of being in an all fours position on a bed.

Experiment 1: Pose Keyword Comparison (Fixed Front Angle)

Condition A: all fours + front (3 tokens)

Observations

On the bed in a near-all-fours forward-leaning posture. However, it’s not a complete all fours — kneeling with hands forward on the bed and kneeling positions are mixed. Pose stability is somewhat low.

Condition B: on all fours + front (4 tokens)

Observations

Same tendency as A. Adding “on” makes no clear difference.

Condition C: on hands and knees + front (5 tokens)

Observations

Equivalent pose to A/B. Using 5 tokens doesn’t improve pose stability.

Condition D: crawling + front (1 token)

Observations

The nuance of a crawling motion is added, with weight shifted forward and arms extended in front. Tends to have a greater forward tilt than A–C.

Experiment 2: Angle Comparison (Fixed all fours)

Condition E: all fours + side angle

Observations

Side angle makes the full-body all fours silhouette easier to read. The back line and leg angle are visible.

Condition F: all fours + rear angle

Observations

All fours from behind. Composition with hips centered in the frame.

Condition G: all fours + low angle

Observations

Low-angle upward-looking composition. Face is close and the body looms overhead.

Note: from below produces a composition where the subject appears to loom over the viewer, giving a strong sense of being looked down upon. The all-fours + low-angle combination is compositionally powerful.

Cross-Condition Comparison

Pose Keyword Comparison (all fours / on all fours / on hands and knees / crawling)

Seed	all fours (A)	on all fours (B)	on hands and knees (C)	crawling (D)
42
123
789

A–C are essentially equivalent. Only crawling has a stronger forward tilt and is slightly more dynamic.

Angle Comparison (front / side / rear / low angle)

Seed	Front (A)	Side (E)	Rear (F)	Low angle (G)
42
123
789

Angle specification has more impact on composition than differences in pose keywords.

Note: Testing 4 pose keywords produced nearly identical results, so the single-token crawling is sufficient. It’s also worth remembering that angle specification has more impact on composition than pose keywords.

Key Findings

Pose Keyword Differences

• all fours / on all fours / on hands and knees produce effectively equivalent output with no difference
• Only crawling is slightly more dynamic (most efficient at 1 token)
• None of them can stably reproduce a complete all fours; sitting and kneeling sometimes appear

Effect of Angle

• Angle specification has more impact on composition than differences in pose keywords
• Side angle (from side): Full-body silhouette visible, best for checking the pose
• Rear (from behind): Hip-centered composition
• Low angle (from below): Overhead-view composition with impact

Recommended Prompts

• Basic: all fours (2 tokens) is sufficient
• Dynamic: crawling (1 token) is most efficient
• Combining with angle specification for compositional control is practical

Related Articles

The Rules of AI Image Generation Prompts | Word Order, Emphasis Syntax, and Negative Prompt Basics

[Verified] Image Generation Prompt Best Practices