Head to head: Google: Nano Banana Pro (Gemini 3 Pro Image) vs OpenAI: GPT-5.4 Image 2
Google: Nano Banana Pro (Gemini 3 Pro Image) vs OpenAI: GPT-5.4 Image 2
By Ryan Merket · Published
An eight-prompt comparison tests two image models on style, composition, constrained palettes, and text rendering. The briefs reward both faithful instruction-following and visual polish.
The saved scores are **63.3 for Google Nano Banana Pro (Gemini 3 Pro Image)** and **63.8 for OpenAI GPT-5.4 Image 2**. The task record is a tie: each model won three prompts, and two were tied. The half-point aggregate gap does not change that saved result; the editorial strength is **Too close to call (not statistical)**. Google’s clearest advantages were prompt-specific: it better handled the blue-hour loft’s materials, the wide-angle atrium’s zigzag stair and framing, and the dusk campsite’s restricted palette. OpenAI took the ukiyo-e scene, the legible multi-line poster, and the anxious curator portrait, with stronger stylistic cues or clearer narrative focus in each. The two ties show why neither side pulls ahead. Both models placed the room’s furniture effectively despite differences in rendering style, and both produced a readable pavilion poster, balancing exact text and framing differently. **Final call: a tie on this task sample.** The scores and task outcomes point to different strengths, not a decisive overall edge; this assessment applies only to the saved task sample.
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 fully captures the ukiyo-e tradition, with a clearly visible fishing boat riding the crest, a dawn sun, and a muted indigo-and-cream palette. Model A has a dramatic stylized wave, but the boat is partly obscured and the dawn setting is less clear. (Second judge pass, order swapped — scores are the average of both: Model B more fully matches the prompt, with a clearly visible fishing boat and crew riding the crest, a muted indigo-and-cream palette, and convincing ukiyo-e details. Mod | Anthropic: Claude Fable 5.1: Model A is a near-copy of Hokusai's Great Wave…
Binding loft materials
A photoreal editorial interior shot of a compact architect’s loft studio at blue hour, 16:9, with five clearly separated design objects on a long workbench: a matte cobalt ceramic vase, a brushed copper desk lamp, an emerald velvet dining chair, a pale ash wood side table, and a translucent amber glass pendant shade; each object must keep its own exact color and material with no swapping or bleeding, surrounded by neutral concrete walls and pinned floor plans, lit by cool window light plus a soft tungsten practical for crisp material contrast.
OpenAI: GPT-6 Luna: Model A better matches the requested blue-hour 16:9 loft scene, with all five materials and colors clearly represented and the amber shade correctly hanging overhead. Model B stages the objects clearly on a workbench, but its square framing and table-top, vessel-like amber shade depart from the prompt. (Second judge pass, order swapped — scores are the average of both: Model B presents all five materials and colors clearly and separately in a polished, blue-hour studio scene, though it misses th | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 editorial f…
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 flat-vector style and clearly places the bed, centered rug, desk beneath the window, and lamp in the front-right area. Model B has a polished composition and similar layout, but its realistic rendered look departs from the requested flat-vector style. (Second judge pass, order swapped — scores are the average of both: Model B places the bed against the left wall, the desk beneath the back-wall window, the round rug near the floor center, and the lamp clearly in the | Anthropic: Claude Fable 5.1: Both place the bed on the left wall, the desk un…
Wide-angle atrium sweep
A dramatic wide-angle photograph of the lobby atrium of the fictional Marrow & Flint Conservatory, shot from a low corner with a 16mm lens, 16:9, emphasizing exaggerated foreground floor tiles, soaring vertical lines, and slight edge distortion; polished terrazzo, a zigzag stair, hanging plants, and a skylight overhead, in refined contemporary architecture style, with clean morning sunlight slanting through mist for depth and sharp detail from front to back.
OpenAI: GPT-6 Luna: Model A more closely matches the requested wide 16:9 low-angle sweep, with exaggerated terrazzo foreground, a clearly zigzagging stair, skylight, hanging plants, and correctly rendered conservatory signage. Model B has striking sunlight and polish, but its framing is nearly square and its stair is straight rather than zigzagging. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested 16:9 low-angle sweep and features a clear zigzag stair, soaring | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 wide-angle…
Restricted palette
A flat-vector illustration of a mountain campsite at dusk using ONLY four colors — cream, burnt orange, teal, and charcoal — with no gradients and no stray off-palette colors, balanced composition, 16:9.
OpenAI: GPT-6 Luna: Model A more fully matches the dusk campsite prompt, including a moonlit sky, campers, and a campfire, while keeping a cohesive restricted palette and wide composition. Model B is a polished vector scene, but its square format, gradient-like orange sky, and sunlit feel depart from key constraints. (Second judge pass, order swapped — scores are the average of both: Model B has a clear campsite scene and strong graphic styling, but it is square rather than 16:9 and shows subtle tonal variation des | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 format with…
Readable pavilion poster
A clean graphic design photograph of a poster mounted on a brushed steel wall inside a futuristic transit pavilion, cyberpunk neon accents reflected softly in the metal but not obscuring the typography; the poster must be centered, fully visible, and highly legible, reading exactly: "NEXUS COURT / LEVEL 7 / QUIET ATRIUM" in bold sans-serif uppercase, with smaller readable line below: "Open 06:40–22:10"; minimalist layout, high contrast, even lighting, straight-on composition, 16:9.
OpenAI: GPT-6 Luna: Model A follows the requested 16:9, straight-on presentation and shows the exact wording clearly against softly reflected neon. Model B has a more polished futuristic design, but its square format and busier surroundings depart more from the brief. (Second judge pass, order swapped — scores are the average of both: Model B has a polished cyberpunk pavilion setting and a well-balanced poster with highly legible, correctly rendered text. Model A matches the requested 16:9 framing, but its poster i | Anthropic: Claude Fable 5.1: Model A delivers the required 16:9 frame, a cen…
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 models render all three lines correctly and clearly on a deep navy, subtly textured background. Model B has a more cohesive poster-like layout and stronger typographic character, while Model A’s wider format feels less like a centered event poster. (Second judge pass, order swapped — scores are the average of both: Model B more closely captures the minimalist poster feel with elegant hierarchy, generous spacing, subtle grain, and all three lines rendered correctly. Model A is also clear and | Anthropic: Claude Fable 5.1: Both render all three lines perfectly spelled,…
Anxious curator portrait
A cinematic shallow-bokeh portrait of a museum curator standing in a narrow gallery of restored Art Nouveau interiors, visibly anxious as she notices a hairline crack spreading across a miniature plaster dome she is holding; her expression must read unmistakably as alarmed worry—raised inner brows, tense mouth, focused eyes—shot at eye level on an 85mm lens with soft background blur, moody side lighting, realistic skin texture, and muted jewel-toned styling.
OpenAI: GPT-6 Luna: Model B more clearly shows the curator noticing the crack, with her focused gaze and worried expression directed at the miniature dome; its Art Nouveau gallery and muted jewel-toned styling also fit the prompt well. Model A has strong shallow-bokeh atmosphere and a convincing alarmed expression, but her gaze is less connected to the crack and the setting is less distinctly Art Nouveau. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the requested | Anthropic: Claude Fable 5.1: Model A delivers the required unmistakably alarm…
Matchup powered by OpenRouter.