Head to head: Kling Image vs OpenAI: GPT-5.4 Image 2
Kling Image vs OpenAI: GPT-5.4 Image 2
By Ryan Merket · Published
Eight image-generation tasks put scene direction, exact counts, object relationships and framing under pressure. The results reveal a sharp trade-off between prompt fidelity and widescreen composition.
OpenAI’s GPT-5.4 Image 2 takes the matchup **8–0**, with a 66.7 aggregate score to Kling Image’s 46.3 and a decisive 100% statistical verdict. The gap shows up most clearly when a prompt depends on several details landing together, not just on a strong-looking image. GPT-5.4 Image 2 gets the concrete requirements right more consistently: it produces exactly seven cups and seven moon-shaped pastries, and gives the harbor repair scene clearer, more legible technician roles. It also makes the ukiyo-e fishing boat, dawn sun and Fuji read as a coherent scene, and presents the watchmaker’s chronometer work and hand grips more convincingly. Kling has real strengths, especially in composition. It more often honors the requested 16:9 frame, and its harbor image has a stronger cinematic sweep; its indigo-and-foam treatment is striking in the wave scene, too. But those wins in presentation don’t outweigh weaker prompt execution: its pastry image has eight items, and the boat and repair actions are less clear. GPT-5.4 Image 2’s recurring square crops are a genuine miss, not a detail to wave away. **Final call: OpenAI’s GPT-5.4 Image 2 wins decisively on this slate. Kling Image can make the stronger widescreen composition, but GPT-5.4 Image 2 is more dependable when the brief demands exact objects, counts and relationships.**
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 riding the cresting wave, with a dawn sun and Fuji adding a convincing ukiyo-e composition. Model A has dramatic foam and strong indigo contrast, but its boats are less clearly presented and the scene feels less cohesive. (Second judge pass, order swapped — scores are the average of both: Model B more directly fulfills the scene with a clearly readable fishing boat, a visible dawn sun, and a restrained indigo-and-cream palette in convincing ukiyo-e sty | Anthropic: Claude Fable 5.1: Model B delivers a single, clearly readable fis…
anatomy_hands_watchmaker
Documentary photojournalism style, 16:9 — a veteran watchmaker in a cramped repair shop leans over a green bench mat, clearly visible from head to waist, using both hands to adjust the tiny balance wheel of an antique marine chronometer with fine brass tweezers and a loupe pushed up on his forehead; all ten fingers must be visible and anatomically correct, with natural joints, believable grip, and realistic body proportions. The bench is scattered with tiny screws, a parts tray, and handwritten timing notes. Cool window light from the left mixes with a warm task lamp over the bench, creating crisp detail on the hands and face.
OpenAI: GPT-6 Luna: Model B more clearly shows the loupe pushed up on the forehead, a chronometer movement being adjusted, and both hands with believable grips; Model A better matches the requested 16:9 framing and has a particularly strong documentary atmosphere. Model B’s square crop is a notable composition miss, but its closer match to the task edges it ahead overall. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows the open marine chronometer being adjusted with tw | Anthropic: Claude Fable 5.1: Model A nails the 16:9 framing and mixed lighti…
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 A follows the requested 16:9 format and clearly binds the duck, cube, sphere, and cylinder to their specified colors and roles. Model B is polished and places the duck correctly, but its square framing misses the requested aspect ratio and its cylinder is less clearly behind both objects. (Second judge pass, order swapped — scores are the average of both: Model A follows the requested 16:9 format and places the duck on the green cube, with the red sphere in front-right and the blue cylinde | Anthropic: Claude Fable 5.1: Model B correctly binds every color and places…
multi_subject_scene_harbor_repair
Documentary photojournalism, 16:9 — on a storm-battered harbor pier at dawn, five offshore wind technicians coordinate a rotor repair beside a service vessel: one kneels over an open tool case, two hold a long composite blade brace together, one speaks into a headset while pointing at a tablet, and one steadies a coil of orange safety line. The group must read as one coherent, balanced scene with natural spacing and interaction, gulls and stacked crates in the background, wet planks reflecting pale sunrise light, and believable protective gear in sea-green, rust, and charcoal tones.
OpenAI: GPT-6 Luna: Model B more clearly depicts all five technicians carrying out the specified coordinated roles, with the tool case, tablet, orange line, vessel, gulls, and wet dawn pier all visible. Model A has a stronger widescreen cinematic feel, but its actions are less clearly matched to the prompt and the brace arrangement feels less natural. (Second judge pass, order swapped — scores are the average of both: Model B clearly depicts all five technicians in the requested coordinated roles, with the open too | Anthropic: Claude Fable 5.1: Model A honors the 16:9 format and the sea-gree…
binding_color_material_lab_samples
A hyper-detailed studio product photograph of four industrial sample objects arranged on a gray worktable, each with its own exact color and material: a cobalt blue anodized aluminum gear, a matte ivory ceramic valve handle, a translucent amber glass reagent bottle, and a dark walnut wood instrument case. The objects are separated clearly so their colors and materials do not bleed or swap. Clean side lighting with soft shadows, precise reflections appropriate to each material, shallow depth of field, and a tidy engineering-lab aesthetic.
OpenAI: GPT-6 Luna: Model B presents all four requested objects with clearer separation and more recognizable forms, especially the valve handle and instrument case. Model A has appealing side-lit realism, but its case shape is less faithful and its bottle reads more like a laboratory flask. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the requested four distinct samples, especially with its clearly amber reagent bottle, matte ivory handle, and rectangular walnut | Anthropic: Claude Fable 5.1: Model B renders all four objects as recognizable…
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 is more visually cohesive and richly composed, with a clear campsite-at-dusk scene, though its square format misses the requested 16:9 ratio. Model A matches the aspect ratio and includes the campsite elements, but its prominent horizon gradient and sparse layout weaken the restricted-palette brief. (Second judge pass, order swapped — scores are the average of both: Model A matches the requested 16:9 format and has crisp flat-vector shapes, while Model B offers a more evocative, balanced | Anthropic: Claude Fable 5.1: Model A honors the 16:9 ratio but breaks the co…
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 B shows exactly seven matching cups in a clear, straight row, while Model A mixes cup sizes and has a less regular arrangement. Model A fits the requested wide format better, but Model B is substantially stronger on the exact-count requirement. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly seven nearly identical espresso cups in a clearly separated straight row on white marble, though it misses the requested 16:9 format. Model A is 16:9, but the | Anthropic: Claude Fable 5.1: Model A shows seven cups but they are not identi…
counting_seven_pastries
A crisp overhead bakery editorial shot of exactly seven distinct moon-shaped hand pies cooling on a black wire rack in a commercial kitchen, all fully visible and individually countable: one blueberry with sugar crystals, one apricot glazed, one poppy seed, one blackberry lattice, one savory spinach-feta, one pear-ginger with coarse sanding sugar, and one cherry with tiny vent slits. No extra pastries, no cropped items, no duplicates hidden underneath. Bright morning window light, stainless steel counter, natural food styling.
OpenAI: GPT-6 Luna: Model B shows exactly seven fully visible, individually countable moon-shaped hand pies with distinct fillings in a clean overhead arrangement. Model A is polished but has eight pastries, most are round rather than moon-shaped, and several varieties appear duplicated or mismatched. (Second judge pass, order swapped — scores are the average of both: Model B presents exactly seven fully visible, individually countable moon-shaped hand pies, with toppings and fillings that closely match the request | Anthropic: Claude Fable 5.1: Model A shows eight round open-faced tarts rath…
Matchup powered by OpenRouter.