Head to head: Wan vs OpenAI: GPT-5.4 Image 2
Wan vs OpenAI: GPT-5.4 Image 2
By Ryan Merket · Published
A broad image-model comparison puts perspective, exact prompt adherence, visual style, reflections, and readable text under the microscope. The results reward more than polish: composition and whether a scene gets its details right matter too.
OpenAI’s GPT-5.4 Image 2 takes this matchup by a real margin: **64.7 to 55.9**, with a tentative lean verdict and six task wins to Wan’s one, plus a tie. That’s a decisive overall result, not a sweep of every brief. Wan’s clearest win is the grocery aisle: it delivers the requested 16:9 frame and a clean, centered vanishing point, where OpenAI’s square crop and shelf-end column undercut the composition. But OpenAI is stronger across the more varied challenges, from the ukiyo-e wave and boat to the tidepool survey and the blue-hour crane convoy. In those scenes, its advantage is not just atmosphere; it’s clearer organization and more convincing visual storytelling. Neither model is a precision machine. OpenAI’s tidepool image is easier to count, but still shows about 10 shells instead of 11; Wan overshoots with roughly 13. And in the three-line text test, both render the wording correctly, with the result called a tie. OpenAI’s win, then, comes from doing well across different kinds of image-making—not from flawless instruction following in every task. **Final call: OpenAI: GPT-5.4 Image 2 wins decisively overall; Wan is the better pick for a tightly specified wide-angle perspective scene.**
Perspective & scale
A photorealistic one-point-perspective shot down a long empty library aisle, shelves receding to a single vanishing point, books getting consistently smaller with distance and nothing warped or wrongly sized, warm overhead light, 16:9.
OpenAI: GPT-6 Luna: Model A more closely matches the requested 16:9 one-point-perspective aisle, with a clear central vanishing point, consistently receding shelves, and warm overhead lighting. Model B has a convincing aisle perspective but is square rather than 16:9 and its prominent shelf-end column disrupts the composition. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested 16:9 format and presents a clear, centered one-point perspective with warm overhead li | Anthropic: Claude Fable 5.1: Model A delivers a clean 16:9 one-point perspec…
Mangrove field biologist
A natural-history editorial illustration in highly realistic watercolor-and-ink style of a field biologist kneeling on a mangrove boardwalk at sunrise, 16:9, gently examining a rescued juvenile osprey; the person is shown clearly from head to boots with natural anatomy and believable proportions, wearing rolled khaki sleeves, a sun-faded teal vest, and mud-streaked trousers; both hands are fully visible and anatomically correct, one hand supporting the bird’s body and the other holding a small cloth bandage, with the correct number of fingers and no fused or extra digits; nearby sit a canvas satchel, a metal bird-tagging kit, and a thermos, while warm low-angle light filters through mangrove roots and mist over calm tidal water.
OpenAI: GPT-6 Luna: Model A better matches the requested 16:9 natural-history watercolor-and-ink editorial style, with the biologist shown head to boots and the mangrove setting and field equipment clearly staged. Model B has especially atmospheric sunrise lighting and realistic detail, but its square framing and less clearly rolled sleeves make it a slightly weaker match overall. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested wide watercolor-and-ink editori | Anthropic: Claude Fable 5.1: Model B delivers the requested highly realistic…
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 convincingly evokes a traditional ukiyo-e print, with expressive stylized foam and a fishing boat dramatically riding the cresting wave. Model A follows the muted indigo-and-cream palette closely, but Model B’s composition and woodblock-print character are stronger despite its warmer tones and added details. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows a fishing boat riding the cresting wave and combines the dawn sun, muted indigo-an | Anthropic: Claude Fable 5.1: Model B delivers a more authentic woodblock fee…
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 convincing window and apple reflections across the chrome and polished counter. Model B is also richly lit, but its square framing and the ambiguous extra apple-like reflection in the teapot weaken adherence and clarity. (Second judge pass, order swapped — scores are the average of both: Model A matches the requested 16:9 framing and clearly presents the teapot, water glass, apple, and strong reflections, while Model B has more dram | Anthropic: Claude Fable 5.1: Model B clearly renders the red apple and windo…
Aurora reserve trail sign
A crisp photorealistic night scene, 16:9, of a wooden trailhead kiosk in a boreal wildlife reserve under green aurora and soft moonlight, with a clean modern poster pinned squarely to the board and fully legible text that reads exactly: "NO DRONES / QUIET MARSH / ENTRY 05"; the typography must be easy to read from a normal viewing distance, with sharp letterforms, proper spacing, and no garbled characters; around the sign are frost-dusted reeds, a pair of curious arctic foxes at the edge of the frame, and a lantern casting a gentle amber side light that makes the poster stand out against the cold blue snow.
OpenAI: GPT-6 Luna: Model A better matches the requested 16:9 scene and includes the exact slash-separated wording, with the aurora, lantern, reeds, and two foxes clearly visible. Model B has especially crisp, readable typography and striking lighting, but it omits the slash separators, adds poster illustrations, and uses a square composition. (Second judge pass, order swapped — scores are the average of both: Model A matches the requested 16:9 framing and includes the aurora, moonlit marsh, lantern, reeds, foxes, | Anthropic: Claude Fable 5.1: Model A delivers the required 16:9 frame with fu…
Tidepool shell census
A top-down studio-style macro photograph of a wet black basalt tidepool table arranged for a marine survey, lit with soft diffused daylight and subtle reflections, showing EXACTLY 11 distinct moon snail shells placed apart so each one is individually visible and countable; the shells vary naturally in size and cream-to-lavender patterning, with tiny droplets of seawater, threads of green sea lettuce, and two small barnacle clusters attached to the rock, but no extra shells, shell fragments, or hidden overlaps; the composition should feel scientifically clear and aesthetically elegant.
OpenAI: GPT-6 Luna: Model B presents a clear, well-spaced survey arrangement with two barnacle clusters, though it appears to show 10 shells rather than exactly 11. Model A is visually striking but has substantially more than 11 shells and a less countable layout. (Second judge pass, order swapped — scores are the average of both: Model B has a clear, elegant survey-like layout with realistic wet basalt, sea lettuce, and two barnacle clusters, though it shows 10 shells rather than the required 11. Model A includes | Anthropic: Claude Fable 5.1: Model A shows 13 shells and three barnacle clust…
Salt-flat crane convoy
A cinematic ultra-detailed photograph, 16:9, of a vast white salt flat at blue hour with a dead-straight service road receding to a single clear vanishing point on the horizon; along the road, four unusual wildlife-rescue vehicles are parked at increasing distances in correct scale: nearest a teal six-wheeled crane truck carrying a giant wooden crate marked with painted flamingo silhouettes, then much smaller a yellow amphibious jeep, farther still a red fuel buggy, and near the horizon a tiny silver ATV; to the left of the road stands a colossal rusted observation tower with ladders and platforms, and scattered in the distance are shallow mirror-like pools reflecting the sky and a few faraway flamingos; the perspective must feel geometrically consistent, with no warped road edges or wrongly sized vehicles, crisp long shadows, cool twilight sky, and realistic atmospheric haze.
OpenAI: GPT-6 Luna: Model B better captures the blue-hour atmosphere, reflective pools, rusted tower, and correctly diminishing four-vehicle convoy, with a more cinematic composition. Model A has a clean vanishing point but reverses the intended vehicle scale/order, lacks twilight mood, and places the crane truck awkwardly beside the road. (Second judge pass, order swapped — scores are the average of both: Model B better captures the ordered, diminishing convoy, giant flamingo-marked crate, rusted tower, reflective | Anthropic: Claude Fable 5.1: Model A keeps the 16:9 frame and a dead-straigh…
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: Both render all three lines correctly on a deep navy background, but Model B has a more distinctive, polished poster composition and stronger visual hierarchy. Model A is slightly crisper, while Model B’s grain and condensed headline give it more character without compromising legibility. (Second judge pass, order swapped — scores are the average of both: Model A presents all three lines crisply and accurately in a clean, balanced sans-serif layout that closely matches the minimalist brief. Mode | Anthropic: Claude Fable 5.1: Both Model A and Model B render all three lines…
Matchup powered by OpenRouter.