Head to head: Kling Image vs Google: Nano Banana Pro (Gemini 3 Pro Image)
Kling Image vs Google: Nano Banana Pro (Gemini 3 Pro Image)
By Ryan Merket · Published
A prompt-following test of two image models across spatial relationships, motion, reflections, style, perspective, and explicit exclusions. The results reveal where each model is dependable—and where a polished image still misses the brief.
Google: Nano Banana Pro takes seven of the eight tasks, and the margin is decisive: 65.2 to 51.5, with a tentative lean and no ties. Its advantage is less about prettiness than about putting the requested things in the right places and making the intended scene legible. That shows up in the most demanding compositions. Google put the blue cylinder behind the other objects rather than balancing it on the sphere, made the reactor-core fall read as airborne action instead of a rider holding the object, and built the requested deep gantry perspective around a single vanishing point. It also made the window-and-apple reflections more visible across chrome and water. Google’s wins in named style and layout were similarly concrete: its ukiyo-e scene read as a woodblock print, and its isometric bedroom followed the requested flat-vector look and furniture placement. Attribute-and-color binding was a less tidy comparison—Kling had the stronger frost, wool, and warm/cool-light treatment in one assessment—but Google’s cleaner object arrangement carried the task overall. Kling’s win came on negation, where precision mattered more than atmosphere: its reading nook kept plants, lamps, and artwork out of frame, while greenery visible through Google’s window broke the explicit no-plants rule. That is a real strength, but it is the exception to a pattern of misses across spatial, action, and style instructions. **Verdict: Google: Nano Banana Pro wins decisively—seven task wins to one, with Kling’s best result exposing a useful niche rather than changing the overall matchup.**
Attribute binding
A photorealistic tabletop scene: a matte green cube to the LEFT of a glossy red sphere, a small blue cylinder BEHIND both, and a single yellow rubber duck sitting ON TOP of the cube. Colors must stay on the correct objects. Soft studio light, 16:9.
OpenAI: GPT-6 Luna: Model B places the blue cylinder behind the cube and sphere, matching the requested arrangement; Model A instead balances the cylinder on top of the sphere. Model A has a polished, softly lit look, but Model B follows the spatial relationships more faithfully while keeping the object colors correct. (Second judge pass, order swapped — scores are the average of both: Model B correctly places the matte green cube left of the glossy red sphere, with the blue cylinder behind them and the yellow duck | Anthropic: Claude Fable 5.1: Model A gets colors right but the blue cylinder…
Action & motion
A lean desert courier on an anti-grav sprint board rockets through the collapsing bazaar of Tazir-9 at the exact instant a stolen sapphire reactor core slips from his gloved hand toward a vendor awning below; documentary photojournalism in 16:9, shot low and close with a fast panning camera so the rider’s face and the tumbling core stay sharp while prayer flags, market stalls, and a pursuing security drone streak into directional motion blur; sand kicks up in a curling rooster tail, fabric snaps, one boot lifts off the board for balance, and scattered glass beads arc through the air under harsh late-afternoon sun, creating a frame packed with convincing momentum, instability, and split-second tension.
OpenAI: GPT-6 Luna: Model B more clearly captures the split-second fall of the reactor core toward an awning, with a convincingly airborne board, directional blur, and forceful momentum. Model A is polished and detailed, but the core appears held in the rider’s hand and the board reads as an ordinary wheeled skateboard, making the action less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B better captures the requested split-second chase with a hovering sprint boa | Anthropic: Claude Fable 5.1: Model A uses a wheeled skateboard instead of an…
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 shows the window and red apple reflected across the chrome and in the water, with dramatic side lighting and a strong close-up composition. Model A is polished and convincing, but the required reflections are less distinct, especially in the water. (Second judge pass, order swapped — scores are the average of both: Model B better captures the dramatic side light and shows the window and reddish apple imagery across the chrome and glass, making the reflection requirement feel | Anthropic: Claude Fable 5.1: Model B delivers true dramatic side light with…
Attribute & color binding
On a frost-rimmed command table in the survey vessel Glass Myna, arrange exactly five artifacts in a clean three-quarter studio composition lit by a cool overhead strip and a warm amber monitor glow: a cobalt-blue ceramic orb, a matte black titanium sextant, a translucent emerald-green glass dagger, a brushed copper cube with engraved runes, and a cream-colored wool-bound star atlas tied with a crimson ribbon; every object must keep its own correct color and material with no swapping or bleeding, surfaces should read distinctly as ceramic, titanium, glass, copper, wool, and paper, and the background should remain subdued so the attribute binding is unambiguous.
OpenAI: GPT-6 Luna: Model A more strongly conveys the frost-rimmed table, wool texture, and warm/cool lighting while keeping all five artifacts and their colors distinct. Model B has a cleaner, more evenly spaced layout, but its atlas cover reads as plainer and the frosted setting is less pronounced. (Second judge pass, order swapped — scores are the average of both: Model B presents all five artifacts clearly in a clean three-quarter arrangement, with the blue orb reading more like ceramic and the subdued command- | Anthropic: Claude Fable 5.1: Model B presents all five artifacts cleanly wit…
Named art-style fidelity
A moonlit procession of scholar-mages crosses the lacquered Vermilion Causeway toward the observatory of Saint Rhel, rendered faithfully in ukiyo-e: flat planes of color, elegant contour lines, patterned robes, stylized waves of fog curling around black pine roots, a huge indigo sky with a pale copper eclipse, and a distant tiered pagoda framed asymmetrically at the left; one apprentice carries a brass astrolabe, another a paper lantern painted with silver moths, and the bridge rail repeats in rhythmic geometry; the composition should feel unmistakably like a traditional woodblock print rather than generic anime or painting, with restrained palette, visible print texture, and deliberate negative space.
OpenAI: GPT-6 Luna: Model B more convincingly fulfills the request for a traditional ukiyo-e woodblock print, with patterned robes, visible print texture, stylized fog, and an elegantly rhythmic bridge; Model A includes the eclipse and observatory but feels more like polished anime-inspired illustration. Model B’s main omissions are the eclipse and observatory, while Model A’s modern-looking dome weakens the intended setting and style. (Second judge pass, order swapped — scores are the average of both: Model B more | Anthropic: Claude Fable 5.1: Model B reads unmistakably as a woodblock print…
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 B more clearly matches the requested flat-vector isometric style while placing the bed, centered round rug, desk beneath the back-wall window, and lamp on the right. Model A includes the main furnishings but looks more like a shaded 3D render and has a less clean, more cluttered composition. (Second judge pass, order swapped — scores are the average of both: Model B clearly matches the requested flat-vector isometric style and places the bed, centered rug, desk beneath the back-wall window | Anthropic: Claude Fable 5.1: Model B delivers the requested flat-vector isom…
Perspective & scale
Inside the crater-sized starship drydock of Khepri Station, a documentary-photojournalism style wide shot in 16:9 looks down a long central gantry toward a single clear vanishing point: in the immediate foreground, a mechanic in a saffron pressure suit stands beside a waist-high toolbox and a coil of hose; mid-distance, three suspended cargo skiffs of different lengths hang from numbered cranes 7, 11, and 14; far beyond them, the half-finished exploration cruiser Nacre Spine fills the dock with its bow receding correctly into depth, while tiny welders sparkle along the hull and a maintenance elevator climbs a pillar on the right; cold industrial rim lighting, believable atmospheric haze, straight structural lines, and rigorously accurate relative size so every person, machine, gantry, and ship section feels physically consistent and unwarped.
OpenAI: GPT-6 Luna: Model B more clearly delivers the requested long central gantry and single vanishing point, with the numbered cranes, foreground mechanic and equipment, three skiffs, and maintenance elevator all present. Model A has strong industrial atmosphere and scale, but the cruiser lies across the dock rather than receding into depth and the crane details are less faithful. (Second judge pass, order swapped — scores are the average of both: Model B more closely follows the requested deep central perspecti | Anthropic: Claude Fable 5.1: Model B nails nearly every element: a clear one…
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 includes the requested chair, books, tea, window, and warm afternoon light while keeping plants, lamps, and artwork out of the frame. Model B is an appealing reading nook, but visible greenery through the window violates the explicit no-plants constraint. (Second judge pass, order swapped — scores are the average of both: Model B is warm and inviting, but greenery visible through the window violates the explicit ban on plants. Model A includes the chair, books, tea, and golden window lig | Anthropic: Claude Fable 5.1: Model A satisfies every negation cleanly (no pl…
Matchup powered by OpenRouter.