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
A comparison across eight demanding image-generation tasks, from precise room layouts and poster typography to glass reflections, anatomy, and fast-moving lab scenes. The results turn on instruction-following and composition more than spectacle alone.
OpenAI: GPT-5.4 Image 2 leads 66.3 to 54.9, with four task wins to Kling Image’s two and two ties. The limited confidence makes this a meaningful lean—not a runaway: Kling is competitive, but OpenAI more often gets the requested composition and details right. OpenAI’s strongest wins are concrete. It places the bedroom furniture more faithfully, makes the spiral colony and the subject’s stunned expression easier to read, follows the poster’s exact three-line text layout, and handles the seated epidemiologist’s pose and hands more convincingly. Those are the kinds of prompt-specific requirements that separate a good-looking image from a useful one. Kling takes reflections and glass, where its balanced 16:9 still life beats OpenAI’s less faithful framing, and the cozy-room negation task, where it better matches the requested widescreen scene. But neither model is spotless: the still-life images both miss the apple’s reflection in the water, and the cozy-room outputs both show foliage despite the no-plants instruction. The greenhouse and action scenes split the difference: Kling better follows the requested widescreen compositions, while OpenAI’s outputs remain polished but drift from the framing. **Verdict: OpenAI: GPT-5.4 Image 2 wins on the strength of more reliable layout, text, expression, and anatomy—not by sweeping the field. Kling Image has two clear task wins, but OpenAI’s broader instruction-following earns the 77% lean.**
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 clearly places all requested furniture in the intended locations and presents a polished, cohesive isometric bedroom. Model A includes the key items, but its harsh shadows, sparse styling, and off-center rug make the composition less clean and less faithful to the requested flat-vector look. (Second judge pass, order swapped — scores are the average of both: Model B clearly places the bed along the left wall, a round rug near the center, the desk beneath the back-wall window, and the lam | Anthropic: Claude Fable 5.1: Model B places the bed against the left wall, t…
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 A better matches the requested 16:9 close-up and presents the teapot, glass, apple, and polished counter in a balanced, photorealistic scene. Model B has especially clear window reflections in the chrome, but neither image clearly shows the apple reflected on the water’s surface. (Second judge pass, order swapped — scores are the average of both: Model B has a more dramatic close-up, but it misses the requested 16:9 format and the apple’s reflection in the water is unclear. Model A matches | Anthropic: Claude Fable 5.1: Model A delivers the required 16:9 frame, stron…
Emotion & facial expression
Close portrait of a marine microbiologist realizing a petri dish has revealed an impossible spiral colony pattern, her expression unmistakably stunned awe—eyes widened, brows lifted, lips parted in a small breathless smile—while the dish glows faint turquoise near her face; painterly realism, dark laboratory background falling out of focus, chiaroscuro lighting, intimate composition
OpenAI: GPT-6 Luna: Model B makes the impossible spiral colony unmistakable and pairs it with a wide-eyed, breathless expression, while Model A has a compelling close portrait but its colony pattern is less clearly spiral. Model A has slightly cleaner technical execution, but Model B’s more dynamic composition and stronger storytelling better fulfill the prompt. (Second judge pass, order swapped — scores are the average of both: Model B more clearly captures the requested stunned awe, with widened eyes, lifted brow | Anthropic: Claude Fable 5.1: Model B delivers a clear, dramatic spiral colon…
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 better matches the requested 16:9 framing and clearly includes the chair, book stack, tea, window, and warm light; Model B is also cozy but is square and places the mug atop the books. Both avoid lamps and artwork, though foliage is visible outside each window despite the strict no-plants constraint. (Second judge pass, order swapped — scores are the average of both: Model B is warm and inviting and avoids all prohibited objects, but it is square rather than 16:9 and places the mug on th | Anthropic: Claude Fable 5.1: Both Model A and Model B correctly omit plants,…
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 follows the requested three-line layout with crisp, correctly rendered text and a restrained navy poster aesthetic. Model A is legible and polished, but splits the title across two lines and appears to use a hyphen rather than the requested en dash. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested three-line centered layout, with crisp, correctly rendered text on a deep navy, subtly textured background. Model A is also legible and polished | Anthropic: Claude Fable 5.1: Model B renders exactly three stacked, centered…
Human anatomy & hands
A field epidemiologist seated at a cluttered stainless lab desk, clearly visible from head to knee, calmly threading a microcapillary tube into a portable analyzer with one hand while the other hand holds a labeled sample vial, both hands fully shown with anatomically correct fingers and natural wrist angles, realistic body proportions and posture; editorial scientific illustration in crisp semi-photoreal style, soft window sidelight and subtle rim light, three-quarter view
OpenAI: GPT-6 Luna: Model B better matches the requested head-to-knee seated view and clearly shows both hands holding a sample vial and working at the portable analyzer, with natural posture and a convincing field-lab setting. Model A is polished, but its subject is cropped above the knee and the hand-held pipette does not clearly appear to be threading a microcapillary tube into the analyzer. (Second judge pass, order swapped — scores are the average of both: Model B better matches the seated field epidemiologist | Anthropic: Claude Fable 5.1: Model B fulfills nearly every requirement: a fi…
Negation & exclusion
A pristine xenobotany quarantine greenhouse at dawn, rows of bioluminescent violet pitcher plants under condensation-beaded glass, polished steel benches, irrigation tubes, specimen tags, and a lone amber decontamination lamp reflected in wet floor panels, but absolutely NO people anywhere in the scene; ultra-detailed oil painting, quiet symmetrical composition, cool blue morning light with one warm accent, 16:9
OpenAI: GPT-6 Luna: Model A more closely matches the requested 16:9 dawn scene, with condensation-beaded glass, orderly violet pitcher plants, irrigation hardware, specimen tags, and a single amber lamp reflected on the wet floor. Model B is polished and strongly symmetrical, but its square framing and more prominent quarantine doorway and signage depart slightly from the prompt’s composition. (Second judge pass, order swapped — scores are the average of both: Model A more fully matches the requested 16:9 greenhous | Anthropic: Claude Fable 5.1: Both outputs correctly contain no people, but M…
Action & motion
A cryogenic materials scientist in a cobalt safety apron lunges sideways across a narrow lab aisle to catch a glass ampoule slipping from a magnetic rack, silver droplets of spilled liquid nitrogen exploding into mist around her boots, motion-blurred lab coat hem and flying data sheets conveying real momentum; cinematic photorealism, high-speed shutter feel, dramatic teal-white overhead fluorescence with a warm emergency indicator glowing in the background, low-angle composition, 16:9
OpenAI: GPT-6 Luna: Model A more closely matches the specified widescreen, low-angle lab-aisle scene, with the cobalt apron, reaching catch, flying papers, nitrogen mist, and warm emergency light all clearly present. Model B conveys a dramatic lunge and strong motion, but its square framing and less clearly readable ampoule make it a weaker fit overall. (Second judge pass, order swapped — scores are the average of both: Model A more clearly shows the scientist catching the glass ampoule at the rack and matches the | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 frame, a nar…
Matchup powered by OpenRouter.