Head to head: Wan vs OpenAI: GPT-5.4 Image 2
Wan vs OpenAI: GPT-5.4 Image 2
By Ryan Merket · Published
OpenAI: GPT-5.4 Image 2 takes the matchup on the scorecard, but the task notes reveal a sharper split than the aggregate suggests: OpenAI is stronger at atmosphere and polish, while Wan often sticks closer to the brief’s exact visual constraints.
OpenAI wins overall, 62.7 to 56.0, with a tentative lean verdict and a 5–1–2 task record. Its clearest advantages show up in the tram-shelter scene and room layout, where it adds richer reflections, mood, and finish without losing the core composition. It also makes the cozier reading nook—but neither model obeys the explicit no-plants instruction, and OpenAI misses the requested widescreen format. Wan’s lone win is telling: in the ramen-sign scene, it follows the requested flat-vector editorial style, while OpenAI’s moodier, near-photoreal treatment misses that direction. More broadly, Wan often respects formal constraints better. The task notes for the Kasbah alley and umbrella-counting scene praise Wan’s isometric or widescreen framing and exact count, even though the verdict labels credit OpenAI in both. That mismatch makes the headline tally less persuasive than the confidence figure implies. OpenAI still earns the overall result: its images are often more cohesive and atmospheric, and it leads on the spatial-layout and tram tasks. But this is not a clean demonstration that it follows image prompts better; it is a win for polish, with meaningful deductions for format, style, and instruction misses. **Final call: OpenAI: GPT-5.4 Image 2 wins on the aggregate, but Wan is the more dependable choice when the brief’s explicit constraints matter most.**
Melancholy Ramen Sign Girl
A clean vector illustration of a lone bicycle courier waiting under a flickering ramen shop sign in a narrow side street after light rain, with a wistful late-night mood: simplified geometric buildings, glossy puddle shapes, power lines, a parked vending machine, and a sleeping cat on a crate; use bold flat shapes, precise linework, restrained detail, and a poster-like composition with the courier placed off-center under the sign glow, evoking contemporary editorial illustration rather than realism, 16:9.
OpenAI: GPT-6 Luna: Model A better follows the requested clean vector, flat-shape editorial style and includes the courier, glowing ramen sign, puddles, vending machine, power lines, and cat. Model B has a stronger, moodier composition, but its near-photoreal rendering departs substantially from the prompt’s vector illustration direction; neither image clearly depicts a girl. (Second judge pass, order swapped — scores are the average of both: Model B captures the rainy late-night atmosphere and includes the courier | Anthropic: Claude Fable 5.1: Model A delivers the requested clean flat-vecto…
Neon Tram Shelter Refractions
A rainy pre-dawn street corner in the fictional port district of Varda Quay, shown as a cinematic ultra-detailed photo-real image, 16:9: a curved glass tram shelter with crystal-clear side panels, a chrome ticket kiosk, and a long puddle beside the curb must accurately reflect and refract the scene — including the glowing route sign "N7 Brindle Loop," a cyclist in a saffron rain cape passing behind the glass, the red tram brake lights across the street, and the striped awning of a closed bakery opposite; include realistic distortions through wet glass, mirrored highlights on polished metal, and layered reflections in puddles under cool blue streetlights and a single magenta shop sign.
OpenAI: GPT-6 Luna: Model B more clearly includes the tram’s red brake lights and creates stronger wet-glass distortions, layered puddle reflections, and cinematic pre-dawn atmosphere. Model A matches the requested 16:9 framing and depicts the kiosk and striped awning clearly, but its red lights appear to belong to a car rather than a tram. (Second judge pass, order swapped — scores are the average of both: Model B more fully captures the cinematic pre-dawn mood, with the tram’s red brake lights, cyclist behind rai | Anthropic: Claude Fable 5.1: Model B nails nearly every scene element — corr…
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 B creates the cozier, more complete reading nook, with a steaming mug, books on a side table, and appealing warm light; however, its square format misses the requested 16:9, and foliage is visible through the window in both images despite the no-plants constraint. Model A has the requested wide format, but its books and mug feel less naturally arranged. (Second judge pass, order swapped — scores are the average of both: Model B creates a warm, inviting nook with all the requested furnishin | Anthropic: Claude Fable 5.1: Model A delivers the required 16:9 frame with a…
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: Both images place the bed, centered round rug, desk beneath the window, and floor lamp in the requested locations. Model B has a more refined, cohesive composition and stronger detail, while Model A adheres slightly more closely to the flat-vector style. (Second judge pass, order swapped — scores are the average of both: Model B places the bed, centered rug, desk directly beneath the back-wall window, and floor lamp in the front-right area, with a polished and consistent isometric layout. Model | Anthropic: Claude Fable 5.1: Model B places the desk directly under the windo…
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 A renders all three lines cleanly and correctly in a centered, minimalist layout with subtle grain. Model B is also legible and polished, but its heavier distressed texture and narrow display lettering stray somewhat from the requested crisp, clean sans-serif look. (Second judge pass, order swapped — scores are the average of both: Model B better balances the oversized headline with the two supporting lines, and its subtle grain and deep navy background give the poster a polished feel. Mod | Anthropic: Claude Fable 5.1: Both render all three lines correctly spelled,…
Kasbah Palette Alley
An isometric 3D-rendered night market alley in an old-city street scene, 16:9, lit by hanging lanterns and shop windows, restricted entirely to the named 'Moroccan Kasbah' palette: indigo, terracotta, sand, turquoise, saffron, and ivory — no other colors anywhere; depict stacked spice jars, tiled stoops, woven baskets, a narrow scooter parked beside a tea stall, and painted shutters, all carefully designed so every visible element stays within that exact palette while still feeling richly varied, crisp, and believable.
OpenAI: GPT-6 Luna: Model A better matches the requested isometric 3D night-market scene and 16:9 framing, with the scooter, lanterns, jars, tiled stoops, and stalls clearly represented. Model B is richly atmospheric and detailed, but its conventional perspective and square framing depart from the prompt, and both images stray somewhat from the strict palette. (Second judge pass, order swapped — scores are the average of both: Model B is richly detailed and atmospheric, but it misses the requested isometric 16:9 pr | Anthropic: Claude Fable 5.1: Model A correctly delivers the isometric 16:9 f…
Seventeen Umbrellas at Halden Pike
A clean high-detail street-photography-style image of a windswept pedestrian plaza outside the fictional Halden Pike Station at dusk, 16:9, showing EXACTLY 17 distinct open umbrellas, each fully visible and individually countable, carried by different people crossing the square; no closed umbrellas, no umbrella reflections mistaken for extras, no printed umbrella images on signs, and no occlusion that hides any umbrella edge; include wet paving, a newsstand, tram tracks, and soft sodium-vapor lighting, composed from a slightly elevated angle so every umbrella can be counted unambiguously.
OpenAI: GPT-6 Luna: Model A better matches the requested 16:9 windswept plaza scene and appears to show exactly 17 distinct, countable umbrellas, with wet paving, tram tracks, a newsstand, and clear station signage. Model B is atmospheric and polished, but its square format and apparent 16 umbrellas miss two explicit requirements. (Second judge pass, order swapped — scores are the average of both: Model B has evocative sodium-vapor lighting, clearly visible tram tracks, and well-separated umbrellas, but it is squar | Anthropic: Claude Fable 5.1: Model A delivers the required 16:9 frame and a…
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 matches the requested 16:9 overhead layout and shows exactly seven clearly separated cups in a straight row. Model B is neatly arranged and softly lit, but its square format misses the specified aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B presents exactly seven clearly separated identical cups in a clean, softly lit row, but its square framing misses the requested 16:9 format. Model A also has exactly seven cups and matches the wide format, t | Anthropic: Claude Fable 5.1: Both Model A and Model B render exactly seven s…
Matchup powered by OpenRouter.