💥 Seedream 4 VS. Nano-Banana Hoy toca poner a prueba el nuevo generador Seedream 4 que promete crear imágenes en 4K 😍 La idea fue elegir diferentes retos y situaciones para ponerlo a prueba con Nano Banana 🍌 - Integrando un producto (Monster) - Cambios de ropa + de marca - Imágenes de muy cerca (piel, ojos, boca) - Selfies - Buceando debajo del agua - Expresiones faciales - Imágenes de cuerpo completo para la consistencia del rostro Si se te ocurren más situaciones que no he tenido en cuenta, comenta cuáles y lo pongo a prueba 💕 💌 Y dime... Con cual te quedas de los dos?? 👀

Aria Cruz | Influencer AI

@soy_aria_cruz · Digital creator

INSTAGRAM · 2025-09-11Source

253likes

45comments

Remix This

Recreate with Create/ai Image Generator

Make your own AI viral video

Prompt

[Subject]
A split-panel AI image comparison poster showing two ultra-close human eye crops side by side, each panel focused on a female eye region framed by thin round eyeglasses, dark eyebrow hairs visible in high detail, long curled eyelashes, skin texture and pores preserved at macro range, subtle blush and warm natural complexion, one iris appearing vivid emerald green on the left and the other appearing olive-green to blue-green on the right, editorial benchmark aesthetic, highly controlled side-by-side testing layout, photoreal skin detail, precision comparison design, not a single portrait, not anime, not illustration.

[Environment]
No real environment beyond the face crop itself, the image operates as a visual comparison card or benchmark post, each half isolated in a tall rounded-rectangle panel against a dark teal or blue-green outer background, bottom text labels identifying model outputs, small prompt text centered near the lower middle, zero background narrative, zero room context, zero scenery, the “environment” is the graphic layout of the comparison itself.

[Composition/Camera]
Vertical side-by-side diptych layout, extreme macro crop tightly centered on the eye region, both panels aligned to mirror each other structurally, nose bridge partially visible near the center split, eyeglass rim cutting through the lower eye zone, eyebrow and upper cheek filling the rest of the frame, rounded-corner card styling, social-post benchmark composition, strong bilateral symmetry with meaningful micro-differences, informational visual design rather than expressive portrait framing.

[Lighting]
Soft diffuse frontal light with controlled reflections in the irises, skin rendered evenly without deep shadow, enough highlight to show pore structure and curvature of the glasses, no hard flash, no stylized color gels, realistic beauty-tech demo lighting intended to expose texture fidelity and eye rendering quality fairly across both panels.

[Style/Rendering]
Photoreal AI benchmark poster, ultra-detailed eye macro comparison, clean tech-review social card, model-versus-model evaluation format, realistic skin and lash detail, controlled beauty close-up, graphic presentation layer with typography, modern AI image-generation comparison aesthetic, emphasis on fidelity testing, texture truthfulness, and micro-detail reproducibility.

[Detail constraints]
Preserve the split-panel comparison format.
Keep both eyes close in framing and scale.
Maintain the thin round eyeglass rim in both panels.
Preserve eyebrow density, lashes, under-eye skin texture, and cheek pores.
Keep the iris colors distinct between left and right panels.
Retain the rounded-rectangle card feel with a narrow divider.
Keep bottom label text areas present as part of the visual concept.
Avoid turning this into a normal face portrait.
Avoid fantasy makeup, excessive retouching, or plastic skin.
Preserve the tech-demo / benchmark-post logic.

Negative prompt:
single portrait, anime eyes, doll skin, over-smoothed face, missing glasses rim, broken symmetry, extra eyes, duplicate eyelashes, warped brows, low-detail pores, unrealistic macro blur, textless crop, cluttered background, cinematic scenery, bokeh room, fantasy makeup, neon cyberpunk, painterly texture, watercolor, posterized skin, overexposed highlights, muddy skin tones, weird eye reflections, unreadable comparison layout, distorted face anatomy, AI mush, watermark clutter beyond intended comparison text

Suggested parameters:
Model family: SDXL photoreal macro / realism checkpoint
Aspect ratio: 4:5 or tall social card
CFG / guidance: 5.5-7
Steps: 30-40
Sampler: DPM++ 2M Karras or equivalent
Priority controls: eye realism, pore detail, comparison symmetry, card layout preservation

Delta prompt strategy:
1. If the output collapses into one face, add “side-by-side diptych comparison card, two separate macro eye panels.”
2. If the skin gets over-retouched, add “visible pores, natural under-eye texture, realistic cheek surface detail.”
3. If the glasses disappear, add “thin round eyeglass rim crossing the lower eye region in both panels.”
4. If the comparison feels messy, add “clean benchmark layout, rounded-rectangle panels, narrow central divider.”
5. If the eyes become stylized, add “photoreal human eye macro, natural sclera, realistic iris reflections, documentary fidelity.”
6. If the panels look identical, add “subtle but visible model variation in iris color, texture rendering, and skin grain.”
7. If lighting becomes dramatic, add “soft frontal test lighting, neutral beauty-tech illumination, no cinematic shadows.”
8. If the result loses social-card feel, add “AI model comparison post, bottom labels, prompt caption, review-card presentation.”
9. If the crop becomes too wide, add “extreme close-up macro crop centered on eye, eyebrow, upper cheek, and nose bridge.”
10. If the image drifts into generic ad beauty, add “technical fidelity comparison, benchmark visual, micro-detail evaluation graphic.”

Soy_aria_cruz's Close Up Eye Detail Comparison AI Image

This image is not a portrait in the normal sense. It is a comparison asset. The whole point is to let viewers inspect micro-differences between two model outputs under nearly identical framing. If you prompt it like a beauty close-up, you will miss the actual job of the image.

The visual power here comes from controlled sameness. Both panels are close enough to feel matched, but different enough to invite inspection. That inspection behavior is the product. The image is built to make people zoom, compare, and argue about fidelity.

Why This Works

The first reason is scale. Eye-level macro detail turns tiny rendering differences into big perceptual events. Eyelashes, pore density, under-eye gradients, catchlights, and iris complexity all become reviewable objects instead of background detail.

The second reason is layout restraint. The composition does not waste attention. There is no room decor, no extra styling, and no narrative baggage. Everything is bent toward a single question: which model rendered the close-up better?

The third reason is typography placement. The prompt line and model labels anchor the image as a benchmark card rather than a generic beauty crop. That text is not decoration. It tells the viewer how to read the frame.

Signal Table

Signal	Evidence (from this image)	Mechanism	Replication Action
Benchmark framing	Two matched eye crops side by side	Invites direct model comparison instead of passive viewing	Prompt a diptych or split-panel comparison card explicitly
Micro-detail emphasis	Pores, lashes, eyebrow hairs, iris texture	Makes fidelity differences visible at a glance	Specify “macro eye crop with natural skin texture and lash detail”
Graphic card structure	Rounded panels, divider, bottom labels	Frames the image as an evaluation object	Include “social comparison card with model labels and prompt caption”
Near-symmetry	Both panels share almost identical crop logic	Reduces confounds and improves trust in the comparison	Use mirrored alignment and matched composition language
Color difference cue	Emerald eye left, olive-blue eye right	Creates immediate visible distinction between outputs	Allow one or two controlled differences while preserving layout sameness
Neutral light honesty	Soft frontal light across both panels	Supports fair texture read and avoids theatrical distortion	Use beauty-tech neutral lighting, not cinematic shadow design

Why You Should Not Treat This as a Generic Beauty Prompt

If you describe this only as “close-up eye detail”, the model will likely output a single polished beauty image. That would miss the benchmark structure entirely. The value here is comparative legibility, not standalone glamour.

To preserve that value, the prompt has to name the card logic: two panels, mirrored crop, model-label text blocks, macro inspection purpose, and a restrained test-light setup. Without those constraints, the output becomes aesthetically fine but functionally useless.

Best Use Cases

This structure fits AI model tests, checkpoint comparisons, prompt-engineering educational content, social proof posts for image fidelity, and creator threads where the audience is expected to compare render behavior. It is strong for technical communities because it makes quality differences discussable.

It is a weak fit for portfolio portraits, editorial character work, branding hero images, or emotional storytelling. Comparison cards are analytical by nature. They are designed for inspection, not immersion.

Prompt Technique Breakdown

Prompt chunk	What it controls	Swap ideas (EN, 2-3 options)
split-panel eye macro comparison card	Overall image type and layout logic	diptych benchmark post; side-by-side fidelity card; dual-panel model test graphic
thin round eyeglasses framing the eye	Shared structural cue across both outputs	circular eyewire close-up; metal round frames; eyeglass rim crossing the cheek and eye zone
visible pores, lashes, eyebrow hairs	Micro-detail quality target	skin grain and lash realism; macro facial texture; pore-faithful beauty close-up
rounded-rectangle panels with narrow divider	Social-card UI feel	app-style comparison tiles; rounded benchmark layout; split review card design
bottom model labels and prompt caption	Readability as an evaluation post	named output labels; benchmark footer text; prompt caption overlay
soft neutral frontal lighting	Fairness and texture truthfulness	beauty-tech test light; neutral inspection light; diffuse macro comparison lighting

Three Transfer Recipes

Recipe 1: Turn It into a Skin Test

Keep the diptych layout, but widen the crop to include more cheek, nose, and under-eye area. This is useful when you want to compare pore rendering, blush gradients, and subtle realism rather than just iris detail.

Recipe 2: Turn It into a Glasses Reflection Test

Keep the macro crop but emphasize eyewire, lens reflections, and specular behavior. This works well for benchmarking realism around reflective surfaces and optical edge handling.

Recipe 3: Turn It into a Mascara / Lash Fidelity Test

Reduce iris emphasis and shift the prompt toward individual lashes, lash curl separation, and lid texture. This makes the comparison more cosmetic-detail focused while preserving the same benchmarking structure.

Aesthetic Read

The aesthetic category here is “technical seduction.” It uses beauty imagery to hold attention, but the actual payload is evaluation. The image is attractive enough to stop scrolling, then analytical enough to keep the viewer engaged.

The strongest move is that it refuses background storytelling. That refusal is smart. By stripping away everything except the eye macro and card labels, the image creates a clean laboratory for visual argument.

Execution Playbook

Start by fixing the format. Write “side-by-side macro comparison card” before you describe the face. If you do not establish the card format first, many models will ignore the split-screen logic entirely.

Then define the shared visual anchors: glasses, eyebrow, eyelashes, pores, nose bridge, and eye reflection. These are the elements that must persist across both panels. The differences should be controlled, not random.

After that, add the graphic shell: rounded panels, divider, bottom labels, prompt caption. These details tell the viewer that this is a model comparison artifact rather than a beauty crop accidentally placed next to another beauty crop.

Finally, keep the lighting neutral. The more dramatic the light, the less trustworthy the comparison becomes. Neutral frontal lighting is not boring here. It is what makes the inspection credible.

Quick diagnostic checklist

Before you accept the result, check five things: do both panels still read as matched crops, are pores and lashes visible, is the eyeglass rim intact in both halves, does the divider remain clean, and do the labels still make the image legible as a benchmark card. If any of those fail, the image loses its comparison value.