Head to head: Kling Image vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
Kling Image vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
Eight image-generation prompts put these models through a varied test of counting, negative constraints, typography, style, and reflective materials. The results show how each handles specific instructions when visual polish alone isn’t enough.
Nano Banana 2 takes this matchup decisively: it scored 63.5 to Kling Image’s 52.5, winning six tasks to one, with one tie. The limited confidence result reflects a real gap in consistency across the set—not a clean sweep of every prompt. Google’s strongest results came when the brief demanded a clear, legible target. Its poster text was crisp and correctly formatted, its ukiyo-e fishing scene hewed closely to the requested style, and its cinematic tram poster balanced the subject against the rainy hillside setting. It also got the dumpling count exactly right, though the assortment was still off. Kling’s clearest win was the negative-constraint test: it kept unwanted lamps and artwork out of the reading nook, while Nano Banana let greenery creep into the window view. The glasshouse aperitif prompt was a tie; Kling better captured the smoked coupes and refractions, while Google made the setting and text clearer. Neither model fully solved the reflections-and-glass prompt’s physics challenge. **Verdict: Google: Nano Banana 2 is the stronger image model in this head-to-head. Kling has a notable advantage when exclusions matter, but Google more reliably turns detailed briefs into coherent, on-target images.**
Eleven Dumplings Ledger
A top-down food editorial photograph of an indigo ceramic platter on a walnut counter holding EXACTLY 11 distinct dumplings, all fully visible and individually countable with slight spacing between each piece: 2 pleated soup dumplings, 3 half-moon potstickers, 1 open shumai with orange roe, 2 crystal har gow, 1 spinach-green crescent dumpling, and 2 tiny dessert sesame balls placed beside the platter; sharp natural side light, clean shadows, no extra food items anywhere in frame, 16:9.
OpenAI: GPT-6 Luna: Model B presents exactly 11 clearly separated dumplings, though it appears to have two shumai and only two potstickers rather than the requested quantities. Model A is attractive and well lit, but includes extra dumplings and sesame balls, making the count and assortment less faithful. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly 11 individually visible pieces on an indigo round platter, though it has two shumai and only two potstickers rather than th | Anthropic: Claude Fable 5.1: Model B lands exactly 11 items with the two ses…
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 closely follows the exclusions, with no visible lamps or artwork and only sparse branches outside the window; Model B shows abundant greenery through the window, making the no-plants constraint less successful. Both include the requested chair, books, tea, and warm window light, while Model B has a slightly more layered, cozy arrangement. (Second judge pass, order swapped — scores are the average of both: Model B has a lovely, cozy nook with all requested furnishings, but its window | Anthropic: Claude Fable 5.1: Model A delivers all required elements with dis…
Tidal Taffy Palette
A meticulously styled overhead still life of a seaside confectioner’s tasting board on pale limestone: saltwater taffy twists, a cut blood-orange caramel, a tiny brass candy hammer, striped parchment, and a half-filled cordial glass, rendered as a crisp editorial food photograph in soft noon window light, 16:9, using only the named color palette 'Tidal Taffy' and absolutely no other colors: seafoam green, fog gray, oyster white, rust coral, and ink navy.
OpenAI: GPT-6 Luna: Model A more closely matches the requested overhead tasting board, with recognizable twisted taffy, a cut blood orange, a tiny brass hammer, striped parchment, and a cordial glass. Model B has an appealing arrangement but substitutes wrapped candies and caramel cubes for key requested items, and both images stray from the strict palette with warm gold or brown tones. (Second judge pass, order swapped — scores are the average of both: Model A more clearly delivers the overhead editorial still lif | Anthropic: Claude Fable 5.1: Model B honors the constraints far better: a ge…
Named art style
A ukiyo-e woodblock print of a fishing boat riding a large cresting wave at dawn, faithful to the flat color planes, bold outlines, and stylized foam of the tradition, muted indigo and cream palette.
OpenAI: GPT-6 Luna: Model B more clearly depicts a fishing boat and crew confronting a cresting wave at dawn, with especially faithful ukiyo-e linework, stylized foam, and muted indigo-and-cream colors. Model A is attractive and on-style, but its sailboat and less distinct dawn setting adhere less closely to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B more clearly depicts a fishing boat riding a cresting wave and closely matches the bold outlines, flat muted indigo-and-cr | Anthropic: Claude Fable 5.1: Model B delivers a coherent single fishing boat…
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 crisply and centered, with the requested wording and dash, subtle grain, and a clean minimalist layout. Model A is also legible and correctly spelled, but its heavier distressed texture and hyphen in the time line are less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the minimalist navy poster brief, with crisp, correctly spelled text in the requested centered hierarchy and subtle grain. Model | Anthropic: Claude Fable 5.1: Model A renders all text correctly but has a he…
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 B more clearly delivers a chrome teapot and convincingly reflects the window, while also showing the apple through and reflected in the water glass. Model A has dramatic lighting and strong countertop reflections, but its teapot reads more gold-toned than chrome and the required reflections are less clear. (Second judge pass, order swapped — scores are the average of both: Model B more convincingly matches the close-up chrome-teapot setup, with clear window reflections in the polished meta | Anthropic: Claude Fable 5.1: Both outputs commit the same physics error of r…
Melancholy Ube Tram Poster
A clean stylized vector illustration of a midnight snack tram rattling through a rain-slick hillside neighborhood, its open serving window displaying ube buns, pandan jelly cups, and a steaming enamel kettle, with one tired vendor in a saffron raincoat and a wistful, cinematic mood; flat shapes, elegant linework, subtle grain, poster-design clarity, dramatic streetlamp backlighting, 16:9.
OpenAI: GPT-6 Luna: Model A clearly depicts the requested snacks, steaming kettle, rain, and saffron-clad vendor, but its very close, oversized tram makes the scene feel less like a composed poster. Model B balances the tram with the rain-slick hillside neighborhood, uses elegant linework and dramatic lamplight, and delivers a stronger cinematic poster design while retaining the key food and vendor details. (Second judge pass, order swapped — scores are the average of both: Model B more completely captures the rain | Anthropic: Claude Fable 5.1: Model A nails the flat-vector, grainy poster st…
Glasshouse Aperitif Refractions
Inside a narrow rooftop glasshouse at blue hour, a lacquered café table set for aperitif with two smoked-glass coupes, a bottle of clear yuzu soda, a chrome cocktail shaker, melting ice in a crystal bucket, and a plate of translucent candied citrus; render as a hyperreal scene where every reflective and transparent surface accurately mirrors, refracts, magnifies, and distorts the city garden around it, including window mullions, string lights, plant leaves, and the table edge, viewed from a low three-quarter angle, 16:9.
OpenAI: GPT-6 Luna: Model A more vividly fulfills the smoked-glass coupes, translucent candied citrus, and prominent refractions of the glasshouse surroundings. Model B presents a polished, well-composed scene with clearer setting and text, but its coupes and citrus are less faithful to the requested details. (Second judge pass, order swapped — scores are the average of both: Model A more clearly delivers the smoked-glass coupes, clear soda bottle, low three-quarter view, and striking reflections of lights and glas | Anthropic: Claude Fable 5.1: Model A delivers the requested low three-quarte…
Matchup powered by OpenRouter.