soy_aria_cruz: Nano Banana Crying Face AI Art

Nano Banana 2 Vs. Nano Banana PRO 💥 Google acaba de lanzar un nuevo generador de imágenes... Lleva un 2 pero no significa que sea mejor que el Pro 👀 (No es Nano Banana Pro 2) Para ponerlo realmente a prueba, las imágenes que he seleccionado para testearlo son todas las que Nano Banana Pro me daba "poco realistas" Tras ver los resultados... Sigo pensando que la versión Pro lo hace mejor que la nueva 😅 Pero si es verdad que en algunas ocasiones no es así! Igualmente quiero escuchar tu opinión al respecto 💌 Y comenta "ARIA" si quieres que te pase los prompts de todas las imágenes 💕

Aria Cruz | Influencer AI

@soy_aria_cruz · Digital creator

INSTAGRAM · 2026-02-27Source

954likes

553comments

Remix This

Recreate with AI Image Generator

Make your own AI viral video

Prompt

[Subject] A split-screen comparison image showing two near-identical close-up portraits of the same young woman crying intensely indoors. She appears in her early 20s with light skin, long dark brown to black hair tied into a loose high ponytail or half-up pony, large green-hazel eyes, thick dark brows, round thin silver eyeglasses, large silver hoop earrings, and a heather-gray hoodie with visible drawstrings. In both panels her face is flushed and distressed, mouth open in a trembling cry, brows arched upward, nose red and irritated, and eyes glossy with emotion. The left panel shows a slightly less realistic crying expression with softer tears and flatter skin detail. The right panel shows a more realistic version with visible tear tracks on the cheeks, wet lower eyelids, more believable skin redness, and stronger emotional realism.

[Clothing & materials] Soft fleece-textured gray hoodie with a loose hood, thick drawstrings, and natural sweatshirt folds at the neckline and shoulders. Silver metal-frame round glasses and hoop earrings are consistent in both panels.

[Props/objects] No handheld props. The image includes a dark teal or deep green background border framing both panels. Each portrait is contained inside a rounded-rectangle card with narrow spacing between them. Text labels sit at the bottom of each panel: “NANO-BANANA 2” in yellow on the left and “NANO-BANANA PRO” in white on the right.

[Environment] Blurred domestic bathroom or hallway interior behind each portrait, with soft daylight coming from a window and pale walls/door frames. The background remains secondary and almost identical between panels to emphasize the facial realism comparison.

[Composition] Vertical 4:5 comparison layout split into two tall portrait panels side-by-side. Tight close-up framing from upper chest to above the head, face centered in each panel. Rounded corners on both internal panels and a thin divider gap. Maintain nearly matched pose and framing between left and right so the realism differences are easy to compare. Keep the dark border visible around the comparison.

[Lighting] Soft natural indoor daylight, frontal-to-slight side window illumination, even exposure across both panels, realistic highlights on wet tears and glasses, no flash, no dramatic shadows. The right panel should feel slightly crisper and more dimensional in skin rendering.

[Color palette] Cool neutral home interior tones, gray hoodie, natural skin tones with pink-red irritation around nose and eyes, silver accessories, dark teal frame border, yellow and white text labels. Moderate contrast, realistic color balance.

[Image style] Photorealistic AI model comparison graphic, realistic selfie-like close-up portraits, subtle split-screen editorial testing format, crisp facial detail, emotional expression benchmark image. Not stylized illustration, not meme cartoon, not beauty retouch.

[Subject] Two side-by-side close-up portraits of the same crying young woman wearing round glasses, hoop earrings, and a gray hoodie. Both images show an intense emotional breakdown, but the right side is more realistic with stronger tear tracks, eye moisture, skin redness, and facial texture.
[Environment] Softly blurred indoor home background, likely bathroom or hallway with daylight, kept consistent across both panels.
[Composition/Camera] Vertical 4:5 split-screen comparison with two rounded portrait cards, dark teal outer background, left and right panels aligned symmetrically, bottom labels naming each model version.
[Lighting] Soft natural indoor daylight with realistic reflections on glasses and tears.
[Style/Rendering] Photoreal AI comparison graphic, emotion realism benchmark, close-up selfie portrait style, clean layout for side-by-side model evaluation.
[Detail constraints] Keep exactly two panels of the same woman. Preserve the gray hoodie, glasses, hoop earrings, crying mouth shape, red nose, dark outer border, rounded cards, and bottom labels. Do not add extra text, extra panels, or unrelated design elements.

Negative prompt: extra people, extra panels, collage chaos, beauty makeup glam look, smiling expression, no tears, missing glasses, different hairstyle between panels, wrong hoodie color, split-face distortion, cartoon crying, anime tears, outdoor background, dramatic cinematic shadows, text at top, model watermark, low detail, identical left-right rendering with no realism difference

Suggested parameters:
- Aspect ratio: 4:5
- Lens / focal length: smartphone portrait close-up 35mm equivalent
- Depth of field: shallow-to-moderate, face sharp and background softly blurred
- Steps: 32-42
- CFG / style strength: 6.5-8
- Sampler: DPM++ 2M Karras or equivalent
- Seed suggestion: 193642

Delta prompt strategy:
1. If the comparison layout breaks: “two equal vertical rounded panels side-by-side on a dark teal background”
2. If the emotional expression softens: “intense crying face, open mouth, trembling lips, brows lifted in distress”
3. If the right panel is not more realistic: “right panel with more natural tear tracks, skin moisture, redness, and facial realism”
4. If glasses or earrings disappear: “round silver eyeglasses and medium hoop earrings in both panels”
5. If the hoodie changes: “heather-gray hoodie with fleece texture and visible drawstrings”
6. If the background becomes distracting: “softly blurred indoor hallway or bathroom background, nearly identical in both panels”
7. If the split cards lose polish: “rounded-corner portrait cards with a narrow vertical gap and clean graphic framing”
8. If the tears look fake: “subtle wet tear streaks under the eyes, reflective moisture on lower eyelids”
9. If labels are missing: “bottom text labels reading NANO-BANANA 2 and NANO-BANANA PRO”
10. If both sides look too similar: “left side slightly flatter and less realistic, right side noticeably more lifelike and detailed”

How soy_aria_cruz Made This Nano Banana Crying Face AI Art — and How to Recreate It

Pretty portraits are easy places for image models to hide. Smooth skin, flattering light, and neutral expressions can make weak realism look passable. This comparison does the opposite. It forces the model into one of the hardest human tasks: an emotionally intense crying face in close-up, with glasses, tears, redness, wet reflections, and distorted mouth shape all visible at once. That is why the image is useful.

The split-screen layout also sharpens the judgment. You are not comparing two unrelated outputs. You are looking at the same identity, same framing, same hoodie, same scene, and nearly the same expression. That setup removes excuses. Any difference you notice is more likely to come from the model’s realism handling than from prompt drift.

Why This Test Format Works So Well

Emotion is one of the fastest ways to expose weak image generation. When a model does not fully understand facial strain, crying turns into performance instead of feeling. The mouth shape becomes theatrical, the tears look pasted on, the skin loses believable wetness, or the eyes fail to carry enough tension. In this comparison, the right panel feels stronger because it holds more of those micro-signals together at once.

That is why side-by-side benchmark posts like this get traction. They let the audience participate in judgment immediately. People do not need technical language to understand realism in tears, redness, or eye moisture. The post invites instinctive comparison, which is perfect for comments and debate.

Signal	Evidence (from this image)	Mechanism	Replication Action
High-difficulty expression	Both panels show an intense crying face with open mouth and red nose	Stress-testing emotional realism exposes weak rendering fast	Use hard facial states such as crying, panic, laughter, or disgust when benchmarking realism
Controlled comparison	The same identity, outfit, framing, and interior setup appear on both sides	Controlled variables make the model difference easier to judge	Lock everything except the model version or generation setting
Instant audience participation	Viewers can tell at a glance which side feels more lifelike	Posts that invite judgment naturally earn comments	Use split layouts that encourage viewers to pick a side without explanation
Visible realism micro-signals	The stronger side shows clearer tear paths, wetness, and skin irritation	Micro-texture cues matter more than general “quality” in face realism tests	Prompt for tears, skin flush, eye gloss, and subtle facial asymmetry instead of generic “high realism” language

Where This Kind Of Comparison Fits Best

This format is strongest for AI creator accounts that want to educate, provoke discussion, or prove why a certain model is worth using. It also works well for prompt sellers and workflow educators because the image itself becomes evidence. A clean A/B test is easier to trust than a long explanation about model quality.

Best for model comparison posts: the side-by-side layout gives viewers an obvious way to evaluate outcomes. Change the emotion, but keep the controlled setup.
Best for realism benchmarks: hard facial expressions reveal more than beauty shots. Change the emotion only if it still stresses anatomy and texture.
Best for comment-driven discussions: viewers can instantly argue for left or right. Change labels and framing, not the one-variable-at-a-time logic.
Best for prompt education: it teaches creators what details matter when they ask a model for emotional realism.

It is less ideal for moodboard content, brand storytelling, or purely inspirational feeds. The image is analytical by design, even though it still looks social-first.

Three Transfer Recipes

Laughter realism test. Keep: split-screen, same framing, same outfit, same identity. Change: crying to uncontrollable laughter with tear-filled eyes and wrinkled nose. Slot template: {same scene} {same identity} {hard emotion A/B} {model label comparison}
Fear reaction test. Keep: close-up face benchmark logic and bottom labels. Change: crying to fear or shock with widened eyes, strained brows, and mouth tension. Slot template: {tight portrait setup} {same wardrobe} {fear realism benchmark} {side-by-side model test}
Age-skin realism test. Keep: matched framing and split layout. Change: emotional stress test to mature skin texture, freckles, pores, and subtle wrinkles. Slot template: {comparison layout} {same face} {texture-heavy realism task} {clean A/B labels}

Aesthetic Read

The design is minimal on purpose. Dark outer framing, rounded portrait cards, and small bottom labels keep all attention on the face. That is the right decision for a benchmark post. If the graphic language were louder, it would interfere with the realism judgment. Instead, the layout acts like a lab bench: simple enough that the emotional rendering becomes the whole event.

The gray hoodie is another smart choice. It removes fashion distraction. The glasses and earrings add just enough complexity to make the test harder, but not enough to turn it into a styling post. Everything here is serving one question: which model handles emotional realism better?

Observed	Why It Matters	How To Recreate
Nearly identical pose and framing in both panels	Prevents viewers from confusing prompt drift with model quality	Use the same crop, head position, and clothing on both sides
Red nose, wet eyes, and tear tracks	These are the signals that make crying feel physically believable	Prompt concrete wetness and irritation details, not just “sad face”
Neutral home background	Keeps the test grounded and avoids atmosphere bias	Use a simple indoor setting with soft daylight and blur it slightly
Small bottom labels instead of heavy title text	Lets the audience compare without visual clutter	Place concise model names low in the frame and keep them secondary

Prompt Technique Breakdown

If you want to build useful comparison content, the prompt needs to be more disciplined than expressive. You are not trying to create two different artworks. You are trying to create a controlled evaluation surface. That means matching identity, camera, clothing, and background first, then pushing the difficult realism feature second.

Prompt chunk	What it controls	Swap ideas (EN, 2-3 options)
same woman, same hoodie, same glasses, same close-up framing in both panels	Experimental control	“same profile shot”; “same mirror selfie crop”; “same waist-up framing”
intense crying expression with red nose, glossy eyes, and visible tears	Realism stress test	“laughing until crying”; “panic expression”; “post-workout exhaustion face”
dark border with two rounded comparison cards and bottom model labels	Benchmark presentation	“minimal black background”; “thin white divider”; “clean carousel comparison layout”
soft indoor daylight and blurred home background	Scene neutrality	“bathroom light”; “bedroom window light”; “hallway daylight”
right panel more lifelike with better tear realism	Comparative outcome clarity	“left more stylized, right more realistic”; “A softer, B sharper”; “baseline vs improved realism”
no glam retouch, no beauty polish	Honesty of the test	“natural skin texture”; “visible pores”; “unflattering but real emotion”

How I Would Iterate It

Baseline lock: the split layout, the identity consistency, and the crying realism cues. Those three things define the usefulness of the post. If one drifts, the comparison becomes weaker or more ambiguous than it needs to be.

Run 1: solve the layout and make sure both panels truly match in crop and identity.
Run 2: tune the emotional anatomy: brows, mouth tension, nostril redness, eyelid wetness.
Run 3: separate the realism levels so the stronger side is clearly more believable without becoming a different person.
Run 4: refine the benchmark presentation with labels, border spacing, and cleaner split-card structure.

Quick remix checklist

One hard realism challenge
One controlled A/B layout
One stable identity across both sides
One clear model difference
One audience question they can answer quickly

The broader insight is simple: if you want creators to care about model quality, give them a test where realism can fail in obvious ways. This image does that. It turns emotion into a benchmark, and that makes the comparison both educational and highly commentable.