Key Findings
- All 3 levels produced clearly distinct tear expressions – tear intensity control via prompts is effective
teary eyes, watery eyeswas reproduced in 3/3 images as a moist sheen around the eyes. No tears rolling down the cheeks; just glistening, watery pupilssingle tear rolling down cheekshowed a visible tear streak on the cheek in 2/3 images. One image stayed at the “misty eyes” levelcrying heavily, tears streaming down faceproduced full crying expressions in 3/3 images, with facial distortion, redness around the nose, and multiple tear streaks- As tear intensity increases, hair dishevelment and skin wetness also intensify as a correlated effect
What You’ll Learn
- How well tear intensity can be controlled through prompts
- The visual differences between “misty eyes,” “single tear,” and “heavy crying”
- How tear expressions affect the overall mood of the face
Experiment Setup
| Parameter | Value |
|---|---|
| Model | z-image-turbo (6B, photorealistic distilled model) |
| Steps | 8 |
| Sampler | euler |
| Scheduler | ddim_uniform |
| CFG | 1.0 |
| Image Size | 1024x1024 |
| Seed | 3 fixed seeds per condition |
Base Prompt
Only the {VARIABLE} portion is changed to compare tear intensity across expressions.
Condition A: teary eyes, watery eyes (Misty Eyes)
| Seed 5100 | Seed 5101 | Seed 5102 |
|---|---|---|
![]() | ![]() | ![]() |
Observations
All 3/3 images showed a moist sheen around the eyes. The pupils appear glistening and reflective, giving a “just about to cry” impression. No tears are flowing down the cheeks – it remains at the “tears welling up” stage. The facial expression itself is relatively calm, with the mouth either closed or slightly parted. The skin surface also shows a dewy, wet-looking texture, as if dampened by sweat or tears.
Condition B: single tear rolling down cheek (One Tear)
| Seed 5110 | Seed 5111 | Seed 5112 |
|---|---|---|
![]() | ![]() | ![]() |
Observations
2/3 images (seed5110, seed5111) showed a single tear streak on the cheek. Seed5112 stayed closer to the misty-eyed level with wet skin texture rather than a distinct tear, making it difficult to distinguish from Condition A. In the two images where tears were rendered, a thin tear line is visible running from the inner corner of the eye down the cheek. The expression is slightly more emotional than Condition A, with the eyebrows drawn together. Seed5111 generated a pink ribbon hair accessory not specified in the prompt.
Condition C: crying heavily, tears streaming down face (Sobbing)
| Seed 5120 | Seed 5121 | Seed 5122 |
|---|---|---|
![]() | ![]() | ![]() |
Observations
All 3/3 images clearly reproduced the heavy crying expression. The difference from Conditions A and B is the most dramatic. Specific features observed:
- Tears: Multiple tear streaks running down both cheeks, with visibly greater volume
- Expression: Eyebrows significantly distorted, mouth open in a crying face (3/3 images)
- Redness around nose and eyes: Flushing and redness from crying visible around the nose and eye area (especially prominent in seed5120)
- Disheveled hair: All 3 images show messy hair sticking to the forehead and cheeks from sweat and tears
- Composition shift: Seed5121 shows a forward-leaning posture, and seed5122 shows a lying-down position. Despite the
close-up facespecification, the framing shifted to bust-up or upper body. The intense emotional expression of heavy crying may have influenced the composition as well
Cross-Comparison
| Condition | Tear Volume | Facial Distortion | Reproduction Rate |
|---|---|---|---|
| A: teary eyes | Moist sheen only | Minimal | 3/3 |
| B: single tear | One streak to moist | Mild (brow tension) | 2/3 |
| C: crying heavily | Multiple streaks | Major (open mouth, distorted brows) | 3/3 |
Condition B’s “single tear” is a delicate specification, but it only achieved a 2/3 reproduction rate. Fine-grained control over tear quantity is not entirely stable. Meanwhile, Condition A’s “misty eyes” and Condition C’s “sobbing” are both reproduced relatively consistently.
As tear intensity increases, effects ripple beyond just the facial expression into hair dishevelment, skin wetness, and even composition. When aiming for crying scenes, these correlated effects need to be factored in.
Lab Director’s Comment
Here’s the thing about tears – more isn’t always better. That Condition A “about to cry but holding it back” shimmer hits different. The full sobbing is impressive in its own way, but it’s basically a whole different genre of expression. Honestly kinda bummed that Condition B’s single tear isn’t more consistent. That fragile, fleeting vibe is exactly what I want more of.









![[Experiment] Pain Endurance Expression Prompts — How Far Can Subtle Expression Control Go?](/tips/pain-endurance-expression-test/K_seed42_0_0000_42.webp)
![[Verified] Image Generation Prompt Best Practices](/tips/prompt-best-practices/cover_0_0000_4517457392071889496.webp)


