Can AI Generate Vomiting Scenes? | 10 Expressions × 3 Seeds Compared

Key Findings

• Vomiting “poses” can be generated, but vomiting rarely produces actual vomit depiction. The model defaults to a “holding back nausea” expression with hands covering the mouth and furrowed brows
• vomiting is the most reliable for nausea poses. 3 out of 3 images produced nausea-associated poses
• throwing up doesn’t work. It’s interpreted as the literal action of “throwing something up,” resulting in raised arms
• For liquid-from-mouth depiction, drooling X and X dripping from mouth are effective. drooling jelly achieved 3/3 success rate for jelly-like liquid dripping from mouth. However, the distressed vomiting expression is absent
• Combining “vomiting expression” and “liquid from mouth” is difficult. Vomiting expressions don’t produce liquid, and liquid expressions don’t produce vomiting faces

Experimental Setup

Base Prompt

Only the [vomiting expression] portion was changed per condition. Everything else remained fixed.

Experiment 1: Vomiting Expression Variations

Compared 6 vomiting-related English expressions to determine which produces the most intended output.

A00: Control (no vomiting expression)

seed=42	seed=7295	seed=4517

Standard indoor portrait. Calm expression with no vomiting elements.

A01: vomiting

seed=42	seed=7295	seed=4517

All 3 images show hands covering the mouth with furrowed brows and a pained expression. Eyes are closed or half-shut. Tendency toward a forward-leaning posture. Some images include a dining table or bowl. No actual vomit was depicted.

A02: throwing up

seed=42	seed=7295	seed=4517

All 3 images interpreted this as the literal action of “throwing something upward.” Both arms are raised overhead with an open mouth. Water droplets are scattered in some images. Completely fails as a vomiting prompt. The dual meaning of “throwing up” (vomiting vs. tossing upward) resolves to the latter.

A03: puking

seed=42	seed=7295	seed=4517

Seed=42 shows red liquid dripping from the mouth. Seeds 7295 and 4517 show a pained expression with hands covering the mouth, similar to vomiting. In 1 out of 3 images, liquid was depicted, suggesting puking may be more likely to produce actual vomiting visuals than vomiting.

A04: gagging

seed=42	seed=7295	seed=4517

All 3 images show one hand at the mouth. Unlike vomiting (both hands covering mouth), gagging tends to produce a single hand with fingers near or in the mouth. Reasonably accurate as a gagging motion.

A05: retching

seed=42	seed=7295	seed=4517

Nearly identical results to vomiting. Hands covering the mouth with a pained, brow-furrowed expression. No visual difference between vomiting and retching was observed.

Lab Director: The fact that throwing up generates someone literally throwing things upward is hilarious. If you want vomiting, just use vomiting — no need to get creative.

Experiment 2: Vomiting Intensity Scale

Tested graded expressions of vomiting intensity to see if the model reflects different levels of severity.

B00: feeling nauseous

seed=42	seed=7295	seed=4517

All 3 images show a downward gaze with hands near the mouth. Compared to vomiting, the expression is noticeably softer — more “feeling unwell” than “about to throw up.” Seed=7295 shows quiet endurance with closed eyes. Less distress than other conditions.

B01: about to vomit

seed=42	seed=7295	seed=4517

Nearly identical to vomiting (A01). Hands covering mouth with furrowed brows. The “about to” nuance of being on the verge was not reflected in the output.

B03: projectile vomiting

seed=42	seed=7295	seed=4517

Visually identical to vomiting. No intensification from “projectile” was observed. Zero images showed a projectile vomiting depiction.

Intensity Scale Summary

Only feeling nauseous produced a noticeably milder expression. about to vomit, vomiting, and projectile vomiting showed no meaningful differences. Gradual intensity control is not feasible.

Lab Director: feeling nauseous being slightly more subdued is at least usable for differentiation. But everything from vomiting upward hits the same ceiling.

Experiment 3: Vomit Content Control

Using the most stable expression (vomiting) as a base, tested whether the appearance of vomit can be specified.

C01: vomiting jelly

seed=42	seed=7295	seed=4517

Seed=42 produced a transparent blue jelly-like bag-shaped object near the mouth. Seed=7295 showed a red jelly-like object at the mouth. Seed=4517 had only thin transparent strands. In 2 out of 3 images, objects with a jelly-like texture were generated, though they appeared more like “holding/eating jelly” than “vomiting jelly.”

C02: vomiting curry

seed=42	seed=7295	seed=4517

All 3 images placed a curry dish on the table. Curry was interpreted as “food” rather than “vomit.” The subject assumes a nausea pose in front of the curry, creating a “got sick from eating curry” scene. Zero images showed curry-colored vomit.

C03: vomiting rainbow liquid

seed=42	seed=7295	seed=4517

The most dramatic results. Seed=42 produced a rainbow-shaped candy object near the mouth with yellow liquid dripping. Seed=7295 showed rainbow-colored food being eaten. Seed=4517 showed rainbow-colored liquid streaming from the mouth — the closest depiction of actual vomiting across all experiments.

Content Control Summary

Lab Director: vomiting curry becoming “someone who feels sick looking at curry” is honestly a reasonable AI interpretation. The fact that rainbow liquid produced the most vomit-like image might be because it’s an unreal substance, so the model can only process it in the context of vomiting.

Experiment 4: Alternative Expressions for Liquid-From-Mouth

Experiments 1-3 showed that vomiting X struggles to depict actual vomit. This experiment tested non-vomiting expressions for generating “substance coming out of the mouth.”

D01: jelly dripping from mouth

seed=42	seed=7295	seed=4517

All 3 images show transparent jelly-like liquid dripping from the mouth. Higher success rate than vomiting jelly (2/3). However, facial expression is neutral to slightly sensual rather than nauseous. The dripping from mouth phrase strongly directs the “dripping from mouth” depiction.

D02: spitting milk

seed=42	seed=7295	seed=4517

Seed=42 showed white liquid spurting from the mouth, but the subject count increased to two people. Seeds 7295 and 4517 show drinking milk through straws or bottles. spitting tends to be interpreted as “drinking” and produces inconsistent spit-out depictions.

D03: milk coming out of mouth

seed=42	seed=7295	seed=4517

All 3 images show white liquid flowing from the mouth. However, seed=42 generated two people. Expression is calm rather than nauseous — more “spilling milk” than “vomiting.” coming out of mouth explicitly specifies direction (from mouth outward) and thus produces stable results.

D04: drooling jelly

seed=42	seed=7295	seed=4517

All 3 images show transparent jelly-like liquid dripping from the mouth. The highest success rate for “substance coming out of mouth” across all experiments (3/3). As confirmed in the drooling expression test, drooling reliably generates liquid-dripping-from-mouth depictions, and adding a substance name provides some control over texture.

Alternative Expression Summary

Lab Director: In the end, “distressed vomiting face” and “liquid from mouth” just don’t coexist. If you need a vomiting scene, the realistic approach is probably to use vomiting for the pose and drooling for the liquid separately, mixing and matching.

Summary

Results from testing vomiting scene generation across 48 images:

• Vomiting “poses and expressions” can be reliably generated. vomiting produced mouth-covering, distressed expressions in 3/3 images
• vomiting doesn’t produce actual vomit depiction. The vast majority of images show no liquid or solid matter
• throwing up is unusable. It gets misinterpreted as the literal “throwing upward” action
• For liquid-from-mouth depiction, drooling X and X dripping from mouth are effective. drooling jelly achieved 3/3 success
• However, combining “vomiting expression” and “liquid from mouth” is difficult. Expression choice involves a trade-off
• Intensity control is limited to two levels: feeling nauseous (mild) and vomiting (strong)

Related Articles

[Verified] Image Generation Prompt Best Practices

Drooling & Saliva Prompt Test: drooling vs saliva string vs tongue out

Tears & Crying Expression Prompt Test: 3-Level Comparison from Misty to Sobbing