Head to head: FLUX 2 vs OpenAI: GPT-5.4 Image 2
FLUX 2 vs OpenAI: GPT-5.4 Image 2
By Ryan Merket · Published
A contest between polished, richly detailed scenes and tighter control of framing, object relationships, and constraints. The prompts range from ukiyo-e waves and historic textiles to exact counts and spatial layouts.
OpenAI: GPT-5.4 Image 2 takes the saved verdict, with an aggregate score of **64.9** to FLUX 2’s **56.6**. It won five tasks to FLUX 2’s two, with one tie—a clear result on this task sample, not a claim about statistical significance or performance beyond it. OpenAI’s edge came across the more varied, demanding scene prompts: it won the named ukiyo-e style, courtyard placement puzzle, heirloom textile, and spatial layout tasks, as well as the seed-tray task. The wins weren’t flawless: it showed 12 trays instead of 11, and added detail to the courtyard that made some requested relationships less crisp. Still, its outputs were judged stronger overall on those tasks. FLUX 2’s wins were more targeted. It handled the forbidden-object prompt while preserving the requested wide framing, and its tabletop image placed the objects and colors in the specified relationships. OpenAI’s images in those comparisons were polished, but their near-square framing missed the requested 16:9 composition. The greenhouse-reflections task ended in a tie, with each model better matching different parts of the prompt. **Final call: OpenAI: GPT-5.4 Image 2 wins this saved task sample, while FLUX 2 shows a distinct advantage on prompts where framing and precise constraints are decisive.**
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 a dramatic cresting wave at dawn, with a sunlit horizon, stylized foam, and a cohesive ukiyo-e woodblock look. Model A has strong wave imagery and a fitting palette, but its prominent rigged vessel and extra foreground boat make the scene less focused on the requested fishing boat. (Second judge pass, order swapped — scores are the average of both: Model B more faithfully captures the ukiyo-e style and shows a small fishing boat with people ridi | Anthropic: Claude Fable 5.1: Model B delivers a cohesive ukiyo-e print with…
courtyard placement puzzle
Minimalist line art poster, clean black ink with a few flat sage-green accents on warm white paper, showing a small walled gardening courtyard from a slightly elevated front view; place the objects in these exact spatial relationships: a round stone well is in the center, a wheelbarrow is to the left of the well, a stack of three clay pots is to the right of the well, a tall lemon tree in a square planter is behind the well, a kneeling gardener is in front of the well, a watering can is on top of a low wooden crate, and that crate is between the gardener and the wheelbarrow, while a curled hose lies beneath the watering can’s spout; keep all positions unambiguous and clearly readable.
OpenAI: GPT-6 Luna: Model A follows the requested placements closely and keeps the courtyard compact, clearly readable, and minimalist. Model B includes most of the objects but adds substantial architectural and garden detail, and its hose is less clearly positioned beneath the watering-can spout. (Second judge pass, order swapped — scores are the average of both: Model B makes the requested spatial relationships especially clear, including the centered well, tree directly behind it, and crate between the wheelbarr | Anthropic: Claude Fable 5.1: Model B satisfies nearly every spatial constrai…
eleven seed trays
Editorial product-style photograph, top-down 16:9 view of a potting bench made of pale cedar in soft morning window light: show EXACTLY 11 distinct seed trays, all fully visible and individually countable, arranged in a loose grid with no overlaps; 3 trays contain purple basil sprouts, 2 contain tiny carrot seedlings, 4 contain dark moist soil with neat labeled rows but no sprouts yet, 1 contains scattered perlite and an empty dibber, and 1 contains bright green moss; surrounding objects may include a folded linen glove, a brass mister, and handwritten plant tags, but there must be exactly 11 seed trays total and each tray must be clearly separable at a glance.
OpenAI: GPT-6 Luna: Model B clearly depicts the requested categories and a neatly countable arrangement, though it has 12 trays rather than exactly 11. Model A has fewer clearly visible trays, with the basil and moss appearing outside trays, so it misses more of the core requirements. (Second judge pass, order swapped — scores are the average of both: Model B clearly shows the requested tray contents in a tidy, countable arrangement, though it has 12 trays rather than exactly 11 and is not 16:9. Model A has the req | Anthropic: Claude Fable 5.1: Model A shows only 10 trays with just one purpl…
heirloom textile botanist
Ultra-detailed naturalistic portrait of a botanical conservator seated at a wooden table, repairing a rare 19th-century garden exhibition banner embroidered with climbing beans and moonflowers; emphasize fine texture and detail: the linen weave, frayed thread ends, tiny satin stitches, tarnished brass scissors, paper fibers in seed envelopes, dusty terracotta on a nearby pot of thyme, and the sharply rendered leaf venation of pressed specimens pinned to a board behind her; side-lit by a high north-facing studio window, shallow but precise depth of field, every intricate surface rendered crisp without smudging or melted detail.
OpenAI: GPT-6 Luna: Model B more clearly establishes the banner as a 19th-century garden exhibition piece, with convincing embroidered detail and unusually legible period lettering. Model A has a strong, expansive studio composition, but Model B’s textile, conservation activity, and overall historical atmosphere are more cohesive. (Second judge pass, order swapped — scores are the average of both: Model B more clearly conveys a conservator carefully repairing a historic exhibition banner, with convincing linen text | Anthropic: Claude Fable 5.1: Model B renders a convincing 19th-century exhib…
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 captures the warm, cozy nook in the requested 16:9 framing, with the armchair, stacked books, steaming tea, and window clearly present and none of the forbidden items. Model B is also attractive and follows the object constraints, but its square framing misses the requested aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B is warmly lit and polished, but its square framing misses the requested 16:9 format. Model A better matches the requested wide | Anthropic: Claude Fable 5.1: Both Model A and Model B correctly omit plants,…
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 tabletop arrangement especially well, with the duck clearly atop the green cube and the blue cylinder behind the main objects. Model B also depicts the correct objects and colors, but its nearly square framing misses the requested wide composition. (Second judge pass, order swapped — scores are the average of both: Model A matches the requested 16:9 framing and places the duck, cube, sphere, and rear cylinder correctly. Model B has slightly more polished lighti | Anthropic: Claude Fable 5.1: Model A satisfies every constraint: green cube…
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 A follows the requested clean flat-vector isometric style more closely while placing the bed, centered round rug, desk beneath the back-wall window, and lamp in the front-right corner. Model B is more richly polished, but its realistic rendering departs from the requested flat-vector look. (Second judge pass, order swapped — scores are the average of both: Model B presents the requested layout very clearly, with a centered round rug and the desk and floor lamp placed in the intended areas, | Anthropic: Claude Fable 5.1: Both place the bed on the left wall, the rug ce…
greenhouse rain reflections
Cinematic realistic 16:9 interior of a narrow Victorian greenhouse at blue-hour just after rain, viewed from waist height down the center aisle: a horticulturist in a mustard waxed jacket bends over a tray of variegated calatheas, while a ginger cat sits on a damp stone step near the open door; the scene must include a tall antique mirror leaning against the left wall, fogged glass roof panes with raindrops, a chrome watering can, and a shallow puddle on the slate floor, all reflecting and refracting the greenhouse scene correctly and consistently from their own angles; warm grow-lamps glow inside, cool twilight filters through wet glass, with crisp realistic reflections, transparent panes, and subtle distortions in water and curved metal.
OpenAI: GPT-6 Luna: Model A better matches the requested wide, centered-aisle composition and includes the key figures and props in clear relation to the wet greenhouse setting. Model B has beautiful detail and convincing reflections, but its square framing and prominent foreground horticulturist depart more from the prompt’s 16:9, down-the-aisle view. (Second judge pass, order swapped — scores are the average of both: Model B has richer cinematic detail and a particularly clear puddle reflection, but its square fr | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 cinematic f…
Matchup powered by OpenRouter.