Head to head: Google: Nano Banana 2 (Gemini 3.1 Flash Image) vs Krea 3 Edit
Google: Nano Banana 2 (Gemini 3.1 Flash Image) vs Krea 3 Edit
By Ryan Merket · Published
This image-model matchup spans crisp poster lettering, reflective still lifes, exact object counts, cinematic storytelling, and tightly specified scenes. The results are close enough that the task-by-task tradeoffs matter more than any single showcase image.
Krea 3 Edit takes the saved verdict by a narrow margin: **63.6 to 62.0** in aggregate, with four task wins to Google’s two and two ties. That is a **slight edge (not statistical)**, not a runaway result. Krea’s wins came on legible multi-line text, reflections and glass, the skillet action scene, and the Secession tea poster. Google won exact counting and “Bittersweet Win,” while perspective and scale and “Raincoat of Details” were ties. The split points to different strengths in this set: Krea more often came out ahead on composed, visually specific scenes; Google had the edge on a precisely counted arrangement and a narrative brief. The task outcomes are mixed rather than a clean sweep, and the overall gap is small. **Call: Krea 3 Edit, by a slight edge on the saved task sample; this assessment applies only to that sample.**
Legible multi-line text
A minimalist event poster with three lines of crisp, correctly-spelled text stacked and centered: 'NIGHT MARKET' large on top, 'Fridays · 6–11pm' in the middle, 'Riverside Pier 4' at the bottom, on a deep navy background, clean sans-serif, subtle grain.
OpenAI: GPT-6 Luna: Model B presents all three lines with exact, crisp text and a balanced centered hierarchy that feels especially suited to a minimalist poster. Model A is also legible and faithful, but its heavier type and tighter grouping make the layout feel less refined. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the centered, minimalist poster brief, with all three lines crisp and correctly rendered on a subtly textured navy background. Model A also rende | Anthropic: Claude Fable 5.1: Both render all three lines flawlessly with cor…
Reflections & glass
A photorealistic close-up of a chrome teapot on a polished dark marble counter next to a glass of water; the window and a red apple on the counter must be correctly reflected in BOTH the chrome and the water's surface, dramatic side light, 16:9.
OpenAI: GPT-6 Luna: Model A more clearly features the requested chrome teapot, water glass, apple, polished dark marble, and dramatic side light, with stronger visible reflections across the scene. Model B has an elegant, restrained composition, but the apple and window are less clearly reflected in the water and chrome as requested. (Second judge pass, order swapped — scores are the average of both: Model B better matches the close-up, dramatic side-lit scene, with a convincing window reflection in the chrome and | Anthropic: Claude Fable 5.1: Model B renders the window grid correctly in the…
Exact counting
A clean overhead flat lay on white marble of EXACTLY seven identical espresso cups arranged in a single straight row, each clearly separate and individually countable, even soft lighting, 16:9.
OpenAI: GPT-6 Luna: Model A presents exactly seven clearly separated, identical-looking cups in a straight row on white marble, with a clean overhead view and soft lighting. Model B also has seven cups, but the right-facing handles nearly connect the cups into a chain and its row feels more cramped. (Second judge pass, order swapped — scores are the average of both: Model B and Model A both show exactly seven cups in a straight row on white marble, but Model A gives the cups clearer separation and a more polished, | Anthropic: Claude Fable 5.1: Both Model A and Model B deliver exactly seven i…
Bittersweet Win
A stop-motion claymation close-up of a young pastry chef named Mirelle clutching a dented copper whisk after winning the tiny Harbor Crumb bake-off by one point, her face unmistakably showing tearful relief and exhausted joy—wet lower lashes, trembling smile, flushed cheeks, one eyebrow still tense—standing in a cramped tiled bakery at blue hour, with a lopsided black-cherry mille-feuille and a crooked paper scorecard reading 87 beside her; cinematic shallow depth of field, practical tungsten counter lights mixed with cool window light, intimate eye-level framing, 16:9.
OpenAI: GPT-6 Luna: Model A better captures the blue-hour bakery with cinematic warm-and-cool lighting, the bake-off setting, and the 87-point result, while keeping Mirelle’s bittersweet expression clear. Model B has a stronger close-up and expressive tears, but its setting and lighting feel less specific to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B delivers the stronger intimate close-up and a vividly tearful, relieved expression, while Model A adds the name Mirelle, t | Anthropic: Claude Fable 5.1: Model A delivers the full brief with convincing…
Perspective & scale
A photorealistic one-point-perspective shot down a long empty library aisle, shelves receding to a single vanishing point, books getting consistently smaller with distance and nothing warped or wrongly sized, warm overhead light, 16:9.
OpenAI: GPT-6 Luna: Model A delivers a clear, centered one-point perspective with convincingly diminishing books, warm overhead lighting, and a polished photorealistic feel. Model B also follows the prompt very well, but its darker, more repetitive shelves and floor make the scene slightly less inviting and natural-looking. (Second judge pass, order swapped — scores are the average of both: Model B more precisely delivers the requested long, centered one-point-perspective aisle in a 16:9 frame, with clear scale rec | Anthropic: Claude Fable 5.1: Both deliver clean one-point perspective with c…
Skillet Flare Leap
A dynamic food-action scene in stylized cinematic realism: a street cook in a flour-smudged teal apron jerks a blackened steel skillet upward as twelve cumin-dusted chickpeas, two strips of charred scallion, and a ribbon of orange flame arc through the air above a midnight noodle cart, captured at the peak of motion with believable momentum, flying oil droplets, curling steam, and the cook’s wrist twisting hard; low-angle composition from cart height, neon reflections on wet pavement, strong backlight and side rim light, energetic motion blur on the fastest elements while the face and pan stay sharp, 16:9.
OpenAI: GPT-6 Luna: Model B better captures the low cart-height angle and a clear airborne arc of chickpeas and scallions, with vivid flame, steam, and wet neon reflections. Model A is also cinematic and detailed, but its framing feels less low-angle and the tossing action is somewhat less legible. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the low cart-height view and blackened skillet, with the sharp cook and pan framed against airborne chickpeas, scallions, f | Anthropic: Claude Fable 5.1: Model A delivers a far more convincing cinemati…
Raincoat of Details
An ultra-detailed still life of a market stall display: a folded saffron herringbone wool coat draped beside a frost-dusted bottle of tamarind soda, a heap of rambutan with translucent spines, a scratched brass mechanical citrus press with tiny bolts and engraved measurement marks, and a crate of rain-dark shiso leaves beaded with water; rendered in crisp contemporary studio realism with every weave, fur-like fruit texture, leaf vein, condensation droplet, and machined metal edge sharply resolved, lit by a soft skylight from the left with a narrow silver bounce on the right, three-quarter composition, 16:9.
OpenAI: GPT-6 Luna: Model A includes the full market-stall setting and more clearly shows the frost-dusted soda bottle alongside the detailed press, rambutan, coat, and shiso. Model B has a cleaner studio arrangement and strong fabric and leaf textures, but the bottle lacks the requested frost and the overall scene feels less like a market stall. (Second judge pass, order swapped — scores are the average of both: Model A includes the full market-stall display with more sharply resolved fruit, condensation, leaves, | Anthropic: Claude Fable 5.1: Model A delivers the ultra-detailed brief more f…
Secession Tea Poster
A faithful Vienna Secession poster-style illustration of an elaborate afternoon tea service built around food: a towering pistachio torte, geometric sandwiches, and a silver samovar arranged on a marble table, with two elongated servers in patterned gowns reduced to elegant flat shapes; strict Art Nouveau-era ornament, gold grid motifs, black linear contours, muted cream, olive, plum, and metallic accents, symmetrical composition with decorative borders and stylized lettering for the fictional tea house 'Salon Vespertil', flat graphic lighting, 16:9.
OpenAI: GPT-6 Luna: Model A more fully captures the prompt with a prominent towering torte, geometric sandwiches, ornate samovar, elongated servers, and rich Secession ornament; its main weakness is the garbled lower lettering. Model B has a clean symmetrical layout and readable title, but the food feels smaller and the overall design is less richly ornamented. (Second judge pass, order swapped — scores are the average of both: Model A more fully realizes the Secession poster brief, with a prominent towering torte, | Anthropic: Claude Fable 5.1: Model B is far more faithful to actual Vienna S…
Matchup powered by OpenRouter.