Guía · Abril 2026 · Morfeo Academy
Guía completa de

Cómo dominar Nano Banana y GPT Image 2

nano-banana-vs-gpt.morfeoacademy.com
Compare
Nano Bananavs GPT-2
PromptsGPT Image 2Nano Banana

Un sistema simple para escribir mejores prompts, cuatro ejemplos claros y los 11 prompts exactos que usamos con Nano Banana y GPT Image 2.

La V2 de la guía que nació del carrusel.
Por Paul · @morfeoacademy

Antes de empezar

Tres ideas para escribir mejores prompts.

Las docs oficiales y el test del cavernícola dejan la misma conclusión: describí una escena completa, no una bolsa de keywords.

1. Bloqueá el sujeto

Si el personaje importa, definilo primero: cara, pelo, ropa, energía y lo que no puede derivar.

2. Elegí una cámara

POV, lente, altura o distancia. Mucha obediencia visual se juega en esa línea.

3. Dejá una sola acción

Una imagen, un gesto central. Cuando compiten varias acciones, se cae la composición.

Benchmark: corrimos el mismo prompt para los dos modelos, con el mismo personaje y las mismas referencias. GPT ganó la mayoría cuando importaban anatomía, perspectiva y lógica espacial. Nano respondió mejor en algunas imágenes más limpias o más gráficas.

Sistema simple

Usá este orden.

Sujeto. Cámara. Acción. Después recién afinás entorno y restricciones.

01

Sujeto

Definí quién aparece y qué no puede cambiar: cara, ropa, pelo, edad, energía.

02

Cámara

Elegí tipo de imagen, POV, lente y distancia. No lo dejes implícito.

03

Acción

Describí una sola acción fuerte y cerrá con 1 a 3 riesgos reales: manos, texto, objeto, perspectiva.

Dáselo a tu agente
Act as a senior image prompt engineer.
Turn my rough idea into one production-ready prompt in English.

Rules:
1. Lock the subject first.
2. Define the camera clearly.
3. Keep one main action.
4. Add only the 1 to 3 constraints most likely to fail.

Return:
- Core scene
- Final prompt
- Shorter variant
Template final
Same exact [subject] from the references. Keep [identity traits] identical.
Premium photoreal [image type], never cartoon, no text overlays.
[camera / lens / perspective].
[one main action].
[environment / lighting / materials].
[1 to 3 failure controls].
Ejemplos

Cuatro casos donde se ve rápido la lógica.

Mirá la escena, quedate con la lección y llevate el prompt exacto ahí mismo.

Caso 01 · Fuego

La única excepción tipográfica.

La única slide con texto dentro del render. Funcionó porque la regla fue rígida: una sola palabra, gigante y atrás del personaje.

Macro Tipografía Encendedor

Qué destrabó la imagen: bloquear la tipografía como excepción y controlar físicamente el encendedor.

Ganó: Nano.

Prompt exacto
Same exact caveman from the references. Keep face, beard, hair, fur tunic, body type and deadpan seriousness identical. Premium photoreal editorial absurdity, never cartoon, never parody. Extreme macro in a dim modern kitchen at night, lens almost touching a glossy stone countertop, the caveman half-crawling into frame holding a simple intact disposable lighter inches from his face as if witnessing the birth of the sun, one thumb making a clean steady flame, realistic lighter anatomy, no broken lighter, no warped plastic, no melted parts, warm fire reflection in his eyes, chrome sink and refrigerator bokeh behind him. This is the only slide allowed to include typography: add one single giant lowercase word 'fuego' in extra-bold warm-bone geometric sans serif, perfectly flat and solid, centered across the full width behind the caveman. The layout should read almost like 'fue' on the left of his head and 'go' on the right, with the full word still obvious and legible. The caveman and lighter must be clearly in front of the word so his head, shoulders and hands crop across the letters, but no letter should cover his face or the flame. Keep the word perfectly legible with generous breathing room, no decorative labels, no contour lines, no extra captions, no other text.
Slide comparativa fuego entre Nano Banana y GPT Image 2
La unica slide donde la tipografia entra adentro del render.
Caso 02 · Auto

Si querés locura, nombrala.

La escena subió cuando el prompt dejó de insinuar emoción y pidió directamente manía, lluvia y exaltación.

Lluvia Avenida del Libertador Cara demente

Qué destrabó la imagen: pasar de “auto raro” a una acción precisa con emoción explícita.

Ganó: GPT.

Prompt exacto
Same exact caveman from the references. Keep face, beard, hair, fur tunic and body type identical. Premium photoreal action-comedy campaign image, never cartoon, no text overlays. Generate this from scratch, not based on any previous composition. Hyper-wide hood-level shot on Avenida del Libertador in Buenos Aires during rain, lens near the front bumper, wet asphalt reflecting headlights and city lights, tree-lined median and elegant urban buildings softly visible through the rain, passing traffic bokeh and streetlamps in the distance. The caveman is mounted on top of a compact modern hatchback as if it were a living horse, gripping the roof rails like reins, knees pressed against the doors, heels digging against the metal body, the car moving through the slick avenue. His face must show wild manic exhilaration: eyes opened wide, feral grin, unhinged joy, rain on his face, as if he has just discovered the ultimate beast and is drunk on the power of controlling it. Headlights cutting through the drizzle, cinematic Buenos Aires energy, absurd, intense, majestic, no typography anywhere.
Slide comparativa auto como caballo entre Nano Banana y GPT Image 2
La escena despego cuando la emocion paso a ser parte del brief y no una esperanza.
Caso 03 · Supermercado

El caos bueno necesita control físico.

Esta slide dejó de romperse cuando el prompt empezó a controlar manos, carrito, productos y perspectiva como una foto de campaña real.

24mm Góndola Carrito rebalsado

Qué destrabó la imagen: nombrar con precisión lo que más se suele romper.

Ganó: GPT.

Prompt exacto
Same exact caveman from the references. Keep face, beard, hair, fur tunic, body type and dead-serious survival urgency identical. Premium photoreal editorial absurdity, no text overlays, no parody, no goofy mascot energy. Inside a brightly lit modern supermarket aisle, cinematic 24mm three-quarter angle looking down a long gondola, the caveman caught in true panic-hoarding mode as if he has discovered infinite food and believes it could vanish at any second. He is lunging sideways while clutching a chaotic armful of products against his chest: cereal boxes, jars, bottles, snack bags, pasta, canned goods, all believable and heavy, with one clear anatomically correct hand still grabbing more from the shelf. A single shopping cart beside him is already grotesquely overflowing with supplies, one wheel slightly twisted under the weight, a trail of dropped packages scattered on the shiny tile floor behind him. Shelves packed with color and density on both sides, strong fluorescent ceiling lights, deep aisle perspective, subtle motion in the reach but his face tack sharp. Make it feel like a luxury ad campaign shot inside a supermarket: absurd, frantic, premium, believable. Avoid duplicate limbs, broken fingers, floating products, warped cart anatomy or weird extra people.
Slide comparativa supermercado entre Nano Banana y GPT Image 2
Aca se nota clarisimo lo que cambia cuando el failure control entra de verdad.
Caso 04 · Selfie

Pedí la selfie final, no el aparato.

El salto grande vino cuando el prompt dejó de pedir “un celular con selfie” y pasó a pedir directamente la foto final.

Front camera POV Sin bezel Corrientes

Qué destrabó la imagen: bloquear la lógica de cámara y sacar el teléfono del centro.

Ganó: GPT.

Prompt exacto
Same exact caveman from the references. Keep face, beard, hair, fur tunic and deadpan seriousness identical. Premium photoreal editorial absurdity, no text overlays. This image must look like the final selfie photo itself, not a photo of a phone screen. True front-camera selfie perspective at night on Avenida Corrientes after rain, with the phone body essentially invisible and outside the frame, no visible phone bezel, no visible phone screen, no hand or finger covering the center of the image, no extra objects in his hands. The caveman holds the device awkwardly at arm's length, both forearms entering only from the lower corners with wide-angle distortion, while he studies his own live image with profound spiritual concern and slight existential confusion, lips parted, eyes fixed on himself as if the device were stealing a fragment of his spirit. Cold screen light on his face and fingers, wet asphalt reflections, theater marquees, taxis and neon signage behind him, bizarre, elegant, unmistakably urban.
Slide comparativa selfie entre Nano Banana y GPT Image 2
La diferencia estuvo en pedir la foto final, no la foto del aparato.
Prompts

Los 11 prompts exactos.

Mismo texto para ambos modelos. Abrí el que te sirva, copiá y adaptá.

Slide 02 Fuego
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic, body type and deadpan seriousness identical. Premium photoreal editorial absurdity, never cartoon, never parody. Extreme macro in a dim modern kitchen at night, lens almost touching a glossy stone countertop, the caveman half-crawling into frame holding a simple intact disposable lighter inches from his face as if witnessing the birth of the sun, one thumb making a clean steady flame, realistic lighter anatomy, no broken lighter, no warped plastic, no melted parts, warm fire reflection in his eyes, chrome sink and refrigerator bokeh behind him. This is the only slide allowed to include typography: add one single giant lowercase word 'fuego' in extra-bold warm-bone geometric sans serif, perfectly flat and solid, centered across the full width behind the caveman. The layout should read almost like 'fue' on the left of his head and 'go' on the right, with the full word still obvious and legible. The caveman and lighter must be clearly in front of the word so his head, shoulders and hands crop across the letters, but no letter should cover his face or the flame. Keep the word perfectly legible with generous breathing room, no decorative labels, no contour lines, no extra captions, no other text.
Slide 03 Computadora
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and deadpan intensity identical. Premium photoreal editorial absurdity, no text overlays. Cinematic night office scene inside a glass tower, camera from the point of view of the laptop screen looking outward as if the glowing screen itself were watching him. The caveman sits too close at a sleek desk, eyes locked on the display with ancient suspicion, one dirty fingertip pressing a bright desktop icon while his other hand clutches a chipped stone tool beside the keyboard, charging cable draped like a strange snake, city lights and reflections behind him. The computer should feel like an intelligent glowing slab he is trying to communicate with, luxurious, eerie, believable.
Slide 04 Auto
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and body type identical. Premium photoreal action-comedy campaign image, never cartoon, no text overlays. Generate this from scratch, not based on any previous composition. Hyper-wide hood-level shot on Avenida del Libertador in Buenos Aires during rain, lens near the front bumper, wet asphalt reflecting headlights and city lights, tree-lined median and elegant urban buildings softly visible through the rain, passing traffic bokeh and streetlamps in the distance. The caveman is mounted on top of a compact modern hatchback as if it were a living horse, gripping the roof rails like reins, knees pressed against the doors, heels digging against the metal body, the car moving through the slick avenue. His face must show wild manic exhilaration: eyes opened wide, feral grin, unhinged joy, rain on his face, as if he has just discovered the ultimate beast and is drunk on the power of controlling it. Headlights cutting through the drizzle, cinematic Buenos Aires energy, absurd, intense, majestic, no typography anywhere.
Slide 05 Supermercado
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic, body type and dead-serious survival urgency identical. Premium photoreal editorial absurdity, no text overlays, no parody, no goofy mascot energy. Inside a brightly lit modern supermarket aisle, cinematic 24mm three-quarter angle looking down a long gondola, the caveman caught in true panic-hoarding mode as if he has discovered infinite food and believes it could vanish at any second. He is lunging sideways while clutching a chaotic armful of products against his chest: cereal boxes, jars, bottles, snack bags, pasta, canned goods, all believable and heavy, with one clear anatomically correct hand still grabbing more from the shelf. A single shopping cart beside him is already grotesquely overflowing with supplies, one wheel slightly twisted under the weight, a trail of dropped packages scattered on the shiny tile floor behind him. Shelves packed with color and density on both sides, strong fluorescent ceiling lights, deep aisle perspective, subtle motion in the reach but his face tack sharp. Make it feel like a luxury ad campaign shot inside a supermarket: absurd, frantic, premium, believable. Avoid duplicate limbs, broken fingers, floating products, warped cart anatomy or weird extra people.
Slide 06 Espejo
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and deadpan seriousness identical. Premium photoreal editorial image, no text overlays. Tight shot inside a mirrored elevator, the caveman staring at his own reflection with grave concentration, one palm pressed against the mirror, confused by the duplicate man trapped inside the metal box, chrome walls, ceiling downlights, glowing floor indicator, fur texture against brushed steel, stillness, pressure, elegance.
Slide 07 Burger
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and deadpan seriousness identical. Premium photoreal editorial absurdity, no text overlays. Extreme close-up inside a fluorescent late-night burger joint, the caveman seated at a tiny plastic fast-food table gripping an enormous double cheeseburger with both hands as if it were a hunted beast, caught mid first bite with sacred concentration, molten cheese stretching, sauce dripping down his fingers, crumpled wrappers and fries scattered around him, untouched soda cup nearby, blurred customers and menu glow in the background. The burger must feel absurdly huge but still believable, luxurious food photography mixed with dead-serious prehistoric comedy.
Slide 08 Selfie
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and deadpan seriousness identical. Premium photoreal editorial absurdity, no text overlays. This image must look like the final selfie photo itself, not a photo of a phone screen. True front-camera selfie perspective at night on Avenida Corrientes after rain, with the phone body essentially invisible and outside the frame, no visible phone bezel, no visible phone screen, no hand or finger covering the center of the image, no extra objects in his hands. The caveman holds the device awkwardly at arm's length, both forearms entering only from the lower corners with wide-angle distortion, while he studies his own live image with profound spiritual concern and slight existential confusion, lips parted, eyes fixed on himself as if the device were stealing a fragment of his spirit. Cold screen light on his face and fingers, wet asphalt reflections, theater marquees, taxis and neon signage behind him, bizarre, elegant, unmistakably urban.
Slide 09 Escalera mecanica
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and dead-serious body language identical. Premium photoreal editorial image, no text overlays. Dramatic overhead shot from halfway up a moving escalator inside a modern shopping mall, looking down at the caveman standing alone at the base as he studies the endless mechanical stairs like a dangerous metal river, one foot cautiously testing the first moving step, body fully braced for combat, repeating glass-and-metal geometry, skylight glow, precise comic timing.
Slide 10 Cajero
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and dead-serious expression identical. Premium photoreal editorial absurdity, never cartoon, no text overlays. Steep oblique night shot in a narrow ATM vestibule, camera close to the machine, the caveman leaning in with reverence as a banknote emerges from the slot, one hand hovering beneath it like receiving an oracle gift, the other hand lightly touching the machine as if it were a sacred stone idol, fluorescent bank light above, concrete walls, security camera dome, card-slot glow, no typography anywhere.
Slide 11 Lavadora
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and dead-serious wonder identical. Premium photoreal campaign image, no text overlays. Night laundromat scene, lens low and close to the round glass door of a front-loading washing machine so the spinning circular drum dominates the foreground like a supernatural vortex, the caveman kneeling directly in front of it with sacred concentration, one hand almost touching the glass, colored motion blur reflecting across his face, rows of chrome machines behind him, cold fluorescent light, the machine itself must feel like the portal, no typography anywhere.
Slide 12 Secamanos
Prompt
Same exact caveman from the references. Keep face, beard, hair, fur tunic and deadpan seriousness identical. Premium photoreal editorial absurdity, no text overlays. Tight public restroom shot, the caveman holding both hands under a powerful automatic hand dryer while invisible air blasts his hair and beard backward, expression caught between awe and alarm, white tile, stainless sink, clean modern bathroom lighting, primitive man versus machine wind, absurd, elegant, dead serious.
Fuentes

De dónde sale esta guía.

La base sale de cruzar nuestras corridas con documentación oficial de OpenAI y Google.

OpenAI: sujeto, fondo, estilo, composición, iluminación y contexto. Ver guía.

Google: claridad, especificidad, ejemplos cuando hacen falta e iteración gradual. Prompt design · Image generation docs.

Nuestro benchmark suma la capa práctica: mismo personaje, mismos prompts y comparación manual de obediencia visual.

Seguí por acá

Dos recursos para complementar esta guía.