Head to head: OpenAI: GPT-5.4 Image 2 vs Meta Muse Image Edit
OpenAI: GPT-5.4 Image 2 vs Meta Muse Image Edit
By Ryan Merket · Published
A head-to-head across eight image-editing briefs tests composition, detail, anatomy, and prompt adherence. The results turn on a recurring demand: getting the requested wide frame without losing the scene’s details.
Meta Muse Image Edit wins this matchup on both the aggregate score—67.7 to 60.0—and the task tally, taking seven of eight prompts with one tie. The statistical verdict is decisive, and the examples point to a consistent reason: Muse more often delivered the requested 16:9 composition while keeping the important subjects legible. That edge shows up across very different briefs. Muse made the reading nook wide without adding prohibited decor, rendered the bracelet-making hands clearly, and organized the breakfast still life around its specified color and material pairings. It also won on the Bauhaus pastry poster, the empty-perspective aisle, the ramen macro, and the reflective chrome-and-water close-up—where format and visibility of requested details both mattered. OpenAI’s images were not without strengths: its breakfast scene had a more recognizable moka pot, and its ramen composition showed more of the bowl and toppings. But appealing details did not overcome repeated square or portrait framing when prompts explicitly called for 16:9. The night-market scene was the lone tie, with both models capturing the ensemble and its key props. **Verdict: Meta Muse Image Edit wins decisively.** Across this set, it was the more reliable choice for translating a detailed brief into a wide, readable composition.
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 includes the requested nook elements and avoids all prohibited objects, but its square framing misses the requested 16:9 format. Model B better matches the widescreen composition while keeping the scene cozy, complete, and free of plants, lamps, and artwork. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the requested 16:9 reading nook and includes the chair, stacked books, tea, window, and warm afternoon light without any prohibited items. Mo | Anthropic: Claude Fable 5.1: Model A satisfies all negations and includes ev…
Hands & anatomy
A natural candid photo of a person tying a friendship bracelet, both hands clearly visible with anatomically correct fingers and a plausible grip on the threads, soft daylight, shallow depth of field, 16:9.
OpenAI: GPT-6 Luna: Model B more closely matches the requested 16:9 candid scene, with both anatomically plausible hands clearly gripping the threads and a softly blurred daylight background. Model A is appealing and detailed, but its square framing misses the requested aspect ratio and the bracelet work is less clearly shown. (Second judge pass, order swapped — scores are the average of both: Model A makes the friendship-bracelet activity especially clear, with both hands visibly pinching and guiding the threads a | Anthropic: Claude Fable 5.1: Model B delivers the requested 16:9 frame with…
Color-bound breakfast still life
A meticulously arranged tabletop still life with exact attribute binding: a matte cobalt-blue ceramic teapot, a translucent amber glass honey jar, a brushed copper moka pot, a glossy jade-green lacquered bento box, a folded ivory linen napkin, and a small charcoal-black slate plate holding two apricots; each object must keep its own specified color and material without swapping or bleeding, set on a pale ash-wood table under soft north-window light with clean shadows and a calm editorial composition, 16:9.
OpenAI: GPT-6 Luna: Model B better matches the requested 16:9 editorial layout, with clean soft lighting and all the specified color/material pairings clearly represented. Model A includes a more recognizable moka pot and slate plate, but its square framing misses the requested aspect ratio and its background branch is more distracting. (Second judge pass, order swapped — scores are the average of both: Model B presents the specified objects and color/material pairings in a clean, balanced 16:9 editorial still life | Anthropic: Claude Fable 5.1: Model B delivers the requested 16:9 frame with…
Bauhaus pastry poster
A faithful Bauhaus-style poster illustration for a fictional patisserie named "Kreisel & Crumb," featuring a sliced pistachio torte, a tall cherry cordial glass, and a geometric espresso machine reduced into bold circles, rectangles, grids, and asymmetrical diagonals; limited palette of cream, black, vermilion, mustard, and ultramarine, clean sans-serif lettering, flat graphic planes, precise negative space, sharp edge alignment, and disciplined modernist balance, lit as a studio poster mockup on a plain wall, 16:9.
OpenAI: GPT-6 Luna: Model B better matches the requested 16:9 poster format and presents the name and three featured items in a bold, readable Bauhaus layout. Model A has a more disciplined palette and elegant negative space, but its portrait format departs noticeably from the prompt. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested wide studio poster mockup and includes all three subjects in a more consistently flat, geometric style, with the patisserie name | Anthropic: Claude Fable 5.1: Model A delivers a more disciplined Bauhaus comp…
Night market tea stand ensemble
At a rain-slick midnight food alley, a compact oolong tea stand called "Lantern 47" is run by a woman in a saffron windbreaker pouring steaming tea from a hammered copper kettle while a teenage delivery cyclist in a glossy plum helmet reaches for a paper cup, a sleepy gray whippet curls under the counter beside stacked pomelo crates, and an elderly man in a navy flat cap fans chestnuts at a tiny brazier; red vinyl stools, hanging price tags, drifting steam, reflections in puddles, and a distant tram create one coherent, well-balanced scene, cinematic composition, shallow haze, crisp focal separation, 16:9.
OpenAI: GPT-6 Luna: Model B captures the full ensemble in a cinematic 16:9 alley, with the tea handoff, chestnut brazier, dog, pomelo crates, tram, and wet reflections clearly staged. Model A includes nearly all requested details, but its square framing feels more crowded and less balanced. (Second judge pass, order swapped — scores are the average of both: Model B more fully delivers the requested wide, rain-slick cinematic scene, with all key characters and props clearly visible, strong reflections, and especiall | Anthropic: Claude Fable 5.1: Model A nails nearly every requested element (w…
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 B better matches the requested 16:9 framing while maintaining a convincing, centered one-point perspective, warm lighting, and an empty aisle. Model A also has strong perspective and scale, but its square framing misses the specified aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B delivers a clean, symmetrical one-point-perspective aisle with books and shelves scaling convincingly toward the vanishing point, warm overhead lighting, and the requeste | Anthropic: Claude Fable 5.1: Model A is a 1:1 square rather than the request…
Macro ramen texture study
An extreme close-up of a just-served bowl of black garlic ramen, focusing on crisp fine textures: tightly curled noodles glistening with broth, micro-bubbles of chili oil, fibrous bamboo shoots, sesame-speckled ajitama eggs, peppered slices of chashu showing pores and sear, frayed nori edges, condensation on the bowl rim, and tiny droplets on lacquered chopsticks; razor-sharp macro realism, dramatic side lighting, dark background falloff, no smudging, no painterly blur, 16:9.
OpenAI: GPT-6 Luna: Model B better delivers the requested 16:9 extreme macro with dramatic side lighting, visible droplets on the chopsticks and bowl, chili-oil bubbles, frayed nori, and richly detailed noodles and chashu. Model A includes more of the bowl and toppings, but feels less tightly cropped and misses some of the requested condensation and chopstick details. (Second judge pass, order swapped — scores are the average of both: Model B better captures the extreme macro brief with dramatic side lighting, cris | Anthropic: Claude Fable 5.1: Model B delivers the requested extreme macro wi…
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 B better matches the requested 16:9 close-up and clearly shows the window and apple reflected in the chrome and water. Model A is attractive, but its square framing and less legible reflections make it a weaker fit. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested 16:9 photorealistic close-up, with dramatic side light and prominent window and apple reflections across the chrome, water, and polished counter. Model A includes the key ob | Anthropic: Claude Fable 5.1: Model B honors the 16:9 format, delivers genuin…
Matchup powered by OpenRouter.