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
This matchup tests two image models on constrained illustration, exact object counts, spatial relationships, typography, and action scenes. The prompts reward not just attractive images, but faithful execution of specific visual instructions.
Google: Nano Banana Pro (Gemini 3 Pro Image) takes the saved result, 62.0 to Kling Image’s 53.8, with six task wins to Kling’s two and no ties. The editorial assessment is a clear result on this sample—not a statistical claim—and it applies only to these saved tasks. Nano Banana Pro’s edge came from repeatedly honoring the brief’s particulars: it delivered the requested flat-vector bedroom layout and restricted-palette campsite, bound the duck and colored shapes to the right positions, and captured the sideways vault more convincingly. In the tea-packing scene, its foreground supervisor and visible packing activity made the task clearer, while Kling’s more symmetrical factory view was less specific to tea. Kling had two distinct wins. Its overhead row made all seven espresso cups clearly identifiable, while Nano Banana Pro’s side-on cups obscured the espresso. Kling also kept the shop sign fully in frame with crisp, correctly spelled wording; Nano Banana Pro’s sign was cropped. Both models broke the requested line-count constraint on the sign. The co-op scene is a qualified win for Nano Banana Pro: its roughly 14 visible helmets were closer to the requested 11 than Kling’s roughly 15, but neither image got the count right. **Verdict: Nano Banana Pro is the stronger choice on this saved task sample, chiefly for spatial and action-prompt fidelity; Kling is better when clear overhead counting or fully framed, legible signage matters.**
Restricted palette
A flat-vector illustration of a mountain campsite at dusk using ONLY four colors — cream, burnt orange, teal, and charcoal — with no gradients and no stray off-palette colors, balanced composition, 16:9.
OpenAI: GPT-6 Luna: Model B presents a more complete, balanced campsite scene with crisp vector shapes and a cohesive restricted palette. Model A has a clear dusk-camp setup, but its prominent sky vignette introduces gradients that directly conflict with the prompt. (Second judge pass, order swapped — scores are the average of both: Model B delivers a polished, balanced flat-vector campsite scene with the requested palette and no obvious gradients. Model A has a clear campsite subject, but its pronounced vignette a | Anthropic: Claude Fable 5.1: Model A breaks the key constraint with an obvio…
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 clearly shows exactly seven separate espresso cups in a straight overhead row on white marble, with attractive soft lighting. Model B has seven aligned white cups, but their interiors and espresso are not visible, so they read less clearly as espresso cups and the view does not appear overhead. (Second judge pass, order swapped — scores are the average of both: Model B shows seven cups in a clean row, but the side-on view obscures their interiors and makes them read less clearly as espre | Anthropic: Claude Fable 5.1: Model A delivers exactly seven identical, clear…
Perspective & scale
An ultra-detailed architectural visualization, 16:9, of a specialty tea-packing factory seen from one corner of a long central aisle, with parallel conveyor belts, stacked cartons, forklifts, ceiling trusses, and skylights all receding cleanly to a single vanishing point; a supervisor in the foreground checks a tablet beside waist-high crates while two smaller workers farther down the aisle operate a pallet jack and inspect machinery, all objects correctly scaled by distance so nothing feels warped, with bright neutral industrial lighting and precise straight-line perspective.
OpenAI: GPT-6 Luna: Model B better captures the tea-packing operation, with foreground crates and tablet-checking supervisor, visible tea cartons, and workers using a pallet jack near machinery. Model A has especially clean, symmetrical perspective and strong industrial architecture, but its foreground supervisor is cropped and the tea-specific activity is less evident. (Second judge pass, order swapped — scores are the average of both: Model B more directly matches the requested corner view, with a clear single-po | Anthropic: Claude Fable 5.1: Model A delivers a very clean symmetric one-poi…
Legible text & typography
A realistic street-level photo, 16:9, of a compact locksmith shop at dusk with a freshly painted cream sandwich-board sign on the sidewalk as the clear focal point; the sign must display exactly this easy-to-read text in dark teal sans-serif lettering: "MIRA KEYWORKS" on the first line and "24-MINUTE DUPLICATES" on the second line; warm interior light spilling through the door, a worker at a key-cutting bench in soft background focus, balanced composition, and the text perfectly legible and correctly spelled.
OpenAI: GPT-6 Luna: Model A makes the cream sign the clear, fully visible focal point, with crisp, correctly spelled text and the requested warm shop interior and softly focused worker. Model B also renders the words legibly, but the sign is cropped at the bottom and its composition is less focused; both images break the requested wording across more than two lines. (Second judge pass, order swapped — scores are the average of both: Model A keeps the entire sign in frame and presents the requested wording more clea | Anthropic: Claude Fable 5.1: Both render the text correctly and legibly in d…
Action & motion
A documentary-style cinematic photo, 16:9, of a shipyard welder in a lime safety jacket vaulting sideways across a narrow rain-slick steel beam while a shower of orange sparks erupts from an angle grinder in his outstretched hand, one boot just leaving the beam and a dangling tool tether whipping behind him; low camera near deck level, dramatic diagonal composition, blurred droplets and flying metal dust to emphasize speed, cold blue dawn fog mixed with the warm flare of the sparks, crisp face shield and gloves, convincing mid-action energy.
OpenAI: GPT-6 Luna: Model B captures the requested sideways vault with an airborne boot, outstretched grinder, whipping tether, and a vivid spray of sparks; Model A has the shipyard atmosphere and sparks but reads as a crouched grinding pose rather than a leap. Model B also has the stronger dynamic diagonal composition, while Model A’s wider framing and foggy harbor remain appealing. (Second judge pass, order swapped — scores are the average of both: Model B clearly captures the sideways vault, airborne boot, outst | Anthropic: Claude Fable 5.1: Model B delivers convincing mid-vault energy wi…
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 more clearly places the blue cylinder behind both the cube and sphere while keeping the duck on the cube and the requested colors. Model A is attractively lit, but its cylinder sits beside the cube rather than clearly behind both objects. (Second judge pass, order swapped — scores are the average of both: Model B clearly places the matte green cube left of the glossy red sphere, with the duck on the cube and the blue cylinder behind the objects. Model A has the duck on the cube, but the | Anthropic: Claude Fable 5.1: Model B nails every binding: matte green cube le…
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 follows the requested flat-vector style and clearly places the bed, centered round rug, desk beneath the back-wall window, and lamp in the front-right corner. Model A captures the room layout well but uses a realistic 3D rendering rather than the requested flat-vector style. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the requested flat-vector isometric style and places the bed, centered round rug, desk beneath the back-wall window, and lam | Anthropic: Claude Fable 5.1: Model B delivers the requested flat-vector styl…
Object counting
A vibrant flat-vector design of a neighborhood bicycle repair co-op at opening time, shown in a tidy front-facing workshop interior with exactly 11 distinct red bicycle helmets hanging on a pale mint pegboard wall, each fully visible, separate, and individually countable; mechanics in navy aprons tune bikes below the helmets, with clean shapes, bold outlines, warm morning light through the window, and no extra helmets anywhere else in the scene.
OpenAI: GPT-6 Luna: Model B has a more complete, polished co-op scene and its 14 visible helmets are closer to the requested 11 than Model A’s 15, though neither gets the exact count right. Model A better emphasizes mechanics actively working, while Model B’s composition, details, and readable signage are stronger overall. (Second judge pass, order swapped — scores are the average of both: Model B is closer to the required helmet count, with about 14 visible helmets versus about 15 in Model A, though neither meets | Anthropic: Claude Fable 5.1: Neither hits the required 11: Model A shows roug…
Matchup powered by OpenRouter.