Head to head: Google: Nano Banana 2 (Gemini 3.1 Flash Image) vs Krea 2 Large
Google: Nano Banana 2 (Gemini 3.1 Flash Image) vs Krea 2 Large
By Ryan Merket · Published
Eight image-generation tasks put these models through text rendering, precise counting, scene composition, and prompt constraints. The results reveal a clear difference between specialized strengths and dependable instruction-following.
Google takes the matchup 65.5 to 56.0, winning six tasks to Krea’s one, with one tie. The limited confidence verdict reflects a real gap, not a runaway on every prompt: Krea has a standout result, and the models split the more specialized visual challenges. Nano Banana 2’s advantage is breadth and prompt fidelity. It produced the stronger wide-format funicular, exactly seven overhead espresso cups, and more faithful souk, reading-nook, harbor-crew, and room-layout scenes. Across these prompts, it more consistently honored framing and placed requested details where they belonged—important strengths when an image has to satisfy a brief, not just look appealing. Krea 2 Large earns its win on legible multi-line text, where it cleanly set all three lines on a full-bleed navy background; Nano Banana 2’s distracting pale border undercut its poster. Krea also brought the stronger dew-covered compass macro, but Nano Banana 2’s widescreen framing balanced that result into a tie. Those are meaningful exceptions, not enough to overturn the broader pattern. **Final call: Google’s Nano Banana 2 is the stronger all-around image model here. Choose Krea for polished text-led posters; for varied briefs demanding accurate framing, counts, and scene details, Nano Banana 2 is the more reliable pick.**
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 A has crisp, correctly rendered text and an elegant hierarchy, but the pale mockup border distracts from the requested deep-navy poster. Model B uses the full navy canvas and centers all three legible lines cleanly, though its middle line is somewhat heavier than the supporting text needs to be. (Second judge pass, order swapped — scores are the average of both: Model B presents the exact three lines in crisp, centered type on a full-bleed deep navy, subtly textured background. Model A als | Anthropic: Claude Fable 5.1: Both render all three lines correctly, but Mode…
Glass funicular overlook
Ultra-realistic twilight travel image from inside a glass-walled funicular descending above Valparaíso, where rain-speckled windows, a polished chrome handrail, and a passenger’s round mirror sunglasses all correctly reflect and refract the steep pastel houses, harbor cranes, neon shop signs, and violet sky outside; layered transparency, accurate mirrored geometry, moody cinematic lighting, wide composition, 16:9.
OpenAI: GPT-6 Luna: Model A better matches the requested wide view from inside a glass funicular, with rain-speckled windows, a polished handrail, harbor cranes, and visible city reflections in the passenger’s sunglasses. Model B has compelling twilight color and a steep city view, but its nearly square framing and less distinct funicular/interior details are weaker matches. (Second judge pass, order swapped — scores are the average of both: Model A more clearly establishes the glass funicular interior, with rain-s | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 wide compos…
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 shows exactly seven clearly separate, identical espresso cups in a straight row from an overhead view on white marble, matching the requested wide format. Model B also has seven cups in a row, but the view is not overhead and the image is square rather than 16:9. (Second judge pass, order swapped — scores are the average of both: Model A shows exactly seven separate, identical espresso cups from overhead in a straight row on white marble, matching the requested framing and lighting. Mode | Anthropic: Claude Fable 5.1: Model A shows exactly seven identical, clearly…
Macro dew compass
Macro close-up travel photograph of a weathered brass pocket compass half-buried in moss on the rim of Iceland’s Víti crater, shot with a 100mm macro lens so tiny dew beads cling to the engraved cardinal letters and the background volcanic slope melts into creamy bokeh; cold blue dawn light, razor-thin focus on the compass needle, cinematic natural color, 16:9.
OpenAI: GPT-6 Luna: Model B better delivers the macro dew concept, with conspicuous droplets on the compass and a softly blurred volcanic landscape. Model A has the stronger widescreen framing and crater-like backdrop, but lacks the requested dew beads. (Second judge pass, order swapped — scores are the average of both: Model B better delivers the macro emphasis and striking dew-covered engraved compass against creamy volcanic bokeh, though it misses the requested 16:9 framing. Model A has the correct wide format a | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 frame, a we…
Car-free souk lane
A narrow saffron-and-cobalt alley in the old souk of Chefchaouen at golden hour, with woven baskets, hanging lanterns, potted mint, a sleepy tabby on a doorstep, and two travelers studying hand-painted wayfinding tiles—BUT absolutely no cars, no motorcycles, no bicycles, and no modern traffic signs anywhere; richly textured documentary-style travel image, deep perspective, warm side light, 16:9.
OpenAI: GPT-6 Luna: Model A better captures the saffron-and-cobalt palette, golden-hour side light, baskets, lanterns, potted plants, cat, travelers, and wide deep-perspective lane. Both images replace the requested hand-painted wayfinding tiles with modern-looking signage, but Model B is much darker and misses the 16:9 format. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested 16:9 golden-hour scene, with a strong saffron-and-cobalt contrast and clear market de | Anthropic: Claude Fable 5.1: Model A delivers the 16:9 format, saffron-and-c…
Negation
A cozy reading nook with an armchair, a stack of books, and a mug of tea by a window — with absolutely NO plants, NO lamps, and NO artwork or picture frames anywhere in the frame. Warm afternoon light, 16:9.
OpenAI: GPT-6 Luna: Model A more fully realizes the requested 16:9 reading nook, with the armchair, books, tea, and warm window light clearly arranged, while avoiding the prohibited objects. Model B also includes the key items and avoids plants, lamps, and artwork, but its square framing and darker, less inviting composition fit the prompt less well. (Second judge pass, order swapped — scores are the average of both: Model A creates the fuller, correctly wide reading nook with a clear armchair, books, tea, and warm | Anthropic: Claude Fable 5.1: Model A delivers a genuinely cozy 16:9 nook wit…
Harbor map crew
Vibrant flat-vector travel scene of five people collaborating on a windswept harbor overlook in Tórshavn: a ferry captain kneels to hold open a paper chart, a teenage birdwatcher points toward puffins offshore, a cyclist in a yellow rain cape leans on a loaded touring bike, an elderly woman pours tea from a red thermos onto a stone bench, and a shaggy sheepdog tugs a loose scarf; behind them, turf-roof houses, moored blue boats, and layered North Atlantic hills are arranged into one coherent, balanced composition under crisp overcast light, 16:9.
OpenAI: GPT-6 Luna: Model A closely matches the requested scene, with all five roles clearly staged in a cohesive harbor overlook and the Tórshavn sign rendered correctly. Model B includes most of the requested elements, but its square framing, blank-looking chart, and scarf not being tugged by the dog weaken adherence and composition. (Second judge pass, order swapped — scores are the average of both: Model A more closely matches the requested wide harbor overlook and gives the five-person scene a clear, balanced | Anthropic: Claude Fable 5.1: Model A delivers a true 16:9 flat-vector harbor…
Spatial layout
A clean isometric illustration of a bedroom: a bed against the LEFT wall, a round rug centered on the floor, a desk under the WINDOW on the back wall, and a floor lamp in the FRONT-RIGHT corner. Flat-vector style, consistent perspective.
OpenAI: GPT-6 Luna: Model A more clearly places the window on the back wall and gives the room a polished, consistent isometric layout, though the desk sits somewhat right of the window. Model B includes all the requested furnishings, but its window is on a side wall and the rug feels less centered. (Second judge pass, order swapped — scores are the average of both: Model B more closely follows the requested layout, with the desk directly beneath the back-wall window and a floor lamp in the front-right corner, whil | Anthropic: Claude Fable 5.1: Model A delivers a polished, consistent isometr…
Matchup powered by OpenRouter.