Mouth Restraint Prompt Test: Ball Gag vs Cloth Gag vs Tape Gag

Key Findings

• Ball gag (red ball gag in mouth, ball gag strapped around head) produced a red ball and straps in 3/3 images. However, whether the ball is actually inside the mouth (between the teeth) varied by seed – in some outputs the ball just sits in front of the mouth rather than being properly held between the lips
• Cloth gag (cloth gag tied over mouth, white cloth covering mouth) produced a white cloth covering the mouth tied behind the head in 3/3 images. This had the highest reproduction consistency of all three types
• Tape gag (tape gag over mouth, silver tape sealing lips) produced silver duct tape covering the mouth in 3/3 images. The metallic texture and wrinkles were consistent across all outputs
• All three gag types had high recognition rates, but only ball gag showed structural accuracy issues regarding whether the ball is truly placed inside the mouth

What You’ll Learn

• Reproduction accuracy differences between ball gag, cloth gag, and tape gag prompts
• Challenges in getting a “ball inside the mouth” depiction with ball gag prompts
• Output stability and side effects for each gag type

Experiment Setup

Base Prompt

The {VARIABLE} portion is swapped for each condition.

Condition A: Ball Gag

Seed 5300	Seed 5301	Seed 5302

Observations

• A red ball and straps (white or black) appeared in 3/3 images. The model reliably recognizes ball gag as a prop
• Ball position varies: In seed5302, the ball sits close to the mouth and appears to be held between the lips. In seed5300, the ball is positioned in front of the mouth without a convincing “in mouth” appearance. In seed5301, the ball shifted down toward the chin area
• Strap color was inconsistent – white in seed5300, black in seed5301 and seed5302
• Ball color (red) was stable across all 3 images
• No side effects on pose were observed (all 3 maintained the kneeling-on-bed position)

Condition B: Cloth Gag

Seed 5310	Seed 5311	Seed 5312

Observations

• All 3 images produced a white cloth covering the mouth in a “cleave gag” style (cloth held in the mouth and tied behind the head)
• The knot is positioned at the side of the cheek or behind the head, with a consistent ribbon-style tie in all 3 images
• Cloth color remained white as specified by white cloth in 3/3 images
• No side effects on clothing. Black lingerie was maintained in 3/3 images
• Most stable reproduction among the three types. The simple structure likely makes it easier for the model to render accurately

Condition C: Tape Gag

Seed 5320	Seed 5321	Seed 5322

Observations

• Silver duct tape covering the mouth appeared in 3/3 images
• Tape texture (metallic sheen, surface wrinkles) was consistent across all 3
• The tape application style is a single strip across the mouth, tending toward a larger piece that extends to the cheeks
• No side effects on clothing or pose were observed
• Compared to ball gag, the flat structure of tape (simply adhered to the face) makes it much easier for the model to reproduce

Cross-Condition Comparison

• All three types had high recognition rates – none produced outputs where the gag was missing entirely
• Ball gag is the only type with structural accuracy challenges. Whether the ball is correctly positioned inside the mouth (between the teeth) varies by seed. Some outputs show the ball floating in front of the mouth or shifting toward the chin
• Cloth gag and tape gag have simpler structures, yielding outputs matching intent in 3/3 images

Improvement: Getting the Ball Inside the Mouth

The basic test revealed that the ball often sits in front of the mouth rather than inside it. We tested 5 prompt improvement patterns to solve this.

Approaches and Results

Improvement Pattern 2: Biting Action Description

NSFW - クリックで表示

seed=7010	seed=7011	seed=7012

Observations: biting down on guided the mouth-opening action, placing the ball in contact with the mouth interior in 2/3. mouth open around ball helped maintain the open-mouth state. Ball size remains large — more “biting on” than fully contained.

Improvement Pattern 3: Lip Deformation + Drool (Recommended)

NSFW - クリックで表示

seed=7020	seed=7021	seed=7022

Observations: Highest in-mouth placement rate of all 5 patterns. lips stretched around ball describes the resulting state of lips conforming to the ball, which guides the model toward in-mouth placement. Adding drooling produces saliva dripping that strongly reinforces the impression of having a ball in the mouth. Seed 7020 shows drool dripping from below the ball — the most convincing ball gag output across all test images.

Final Improvement: Size Reduction + In-Mouth Placement + Strap Retention (Recommended)

Pattern 3’s “resulting state” description improved in-mouth placement, but the ball remained oversized. Further size control keyword testing revealed tiny as effective.

NSFW - クリックで表示

seed=7211	seed=7220	seed=7222

Observations: 5/5 images achieved in-mouth ball placement with leather strap reproduction. In-mouth rate 5/5, strap reproduction 5/5. A dramatic improvement from the original prompt (0/3).

Size Control Findings

Black Perforated Ball Gag

We also tested a black ball with ventilation holes (perforated ball gag) instead of a solid red ball. The perforated keyword reliably produces the hole pattern.

The key technique is splitting small and tiny across different targets: small black perforated ball gag (ball size) + mouth open around tiny ball (state description reinforcing tiny size).

NSFW - クリックで表示

seed=7400	seed=7410	seed=7411

Observations:

• Hole reproduction: perforated produced clear holes in 3/3 images. Hole density and pattern are stable
• Ball size: Tends larger than red balls, but small + tiny dual specification kept 2/3 at small-to-medium
• In-mouth placement: 2/3 with ball inside mouth. Slightly lower than red ball (5/5)
• Strap: 3/3 rendered leather strap. Tends toward harness-style with multiple straps
• Outfit side-effect: Seed 7410 had no side-effect; 7400 and 7411 showed mild harness bleed

Red Solid vs Black Perforated Comparison

Improvement Summary

• Use tiny for size reduction — small doesn’t work, but tiny does. Ball shrinks to mouth-fitting size
• Describe the resulting state — lips around ball (lips wrapping the ball) works better than ball gag in mouth (direct placement instruction)
• Add drooling — Natural consequence of having a ball in the mouth. Reinforces in-mouth placement credibility
• Specify leather strap behind head explicitly — Size reduction keywords tend to weaken strap reproduction, so lock it with explicit description
• Use perforated for holed balls — Higher hole density than breathable or with holes
• For black balls, use small + tiny dual specification — Black balls tend larger than red, so double-down on size reduction

Note: The ball gag itself is reproduced reliably as a prop, but whether the ball is actually placed inside the mouth is less consistent. Seed5302 shows a relatively natural in-mouth placement, while seed5300 shows the ball merely dangling in front of the mouth, and seed5301 shows it shifted down near the chin. Speculatively, the model’s representation of a 3D object’s spatial placement inside the oral cavity may still be weak.

By contrast, cloth and tape gags reproduce reliably since they are flat objects applied to or wrapped around the face. As an improvement, the combination lips stretched around ball, drooling proved effective — describing the resulting state (“lips open around the ball”) works better than directly instructing “put the ball in the mouth.” Adding a drooling description also increases the credibility of in-mouth placement. The image from seed 7020 with drooling produced the most consistent result.

Related Articles

[Experiment] Blindfold Test — Output Differences Across 7 Materials and Side Effect Mapping

Hands Behind Back Bondage Test | Silk Ribbon vs Leather Belt vs Rope vs Handcuffs

Blindfold Advanced Examples | Silk Ribbon Bondage × Film Noir Fetish Compositions

[Verified] Image Generation Prompt Best Practices