Head to head: Krea 3 Edit vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
Krea 3 Edit vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
A close image-editing matchup across precise object placement, readable type, stylized scenes and demanding still lifes. The results show where each model is more reliable—and where the differences disappear.
Krea 3 Edit edges the aggregate score, 61.3 to 60.8, but that half-point gap is not a meaningful separation: the evaluation gives just limited confidence that either model is genuinely better. This is a sample tie, not a narrow win to dress up as a verdict. The task results explain the split. Krea was stronger on the neon ramen lane, the isometric bedroom, the glasshouse soda still life and the macro tart, where matching the requested framing and making small visual details legible mattered. Google took attribute binding, the restricted-palette campsite and the bakery alley, with more cohesive scenes and, in the campsite, better discipline around the flat-vector brief. Both handled the multi-line text task well enough to tie. Neither model dominates across the board. Krea’s best work is notably exact about composition and close-up detail; Google’s is often polished and cohesive, but can drift from a specific setup, as in the bedroom’s frontal rather than isometric view. Those are useful tendencies, not a reliable basis for crowning either one overall. **Final call: Tie — Krea 3 Edit and Google’s Nano Banana 2 are effectively even in this matchup.**
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 studio setup closely, with the correct colors, object placements, and duck clearly sitting on top of the cube. Model B is highly polished and photorealistic, but its bright home interior feels less like soft studio lighting and its cube is a paler green. (Second judge pass, order swapped — scores are the average of both: Model B presents all four objects with the requested colors and placements, in a convincing photorealistic tabletop scene with soft studio light. M | Anthropic: Claude Fable 5.1: Both Model A and Model B bind every color corre…
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 presents all three lines of crisp, correctly rendered text in a balanced, centered minimalist layout with subtle grain. Model B also renders the wording correctly, but its heavier typography and large unused space make the composition less refined. (Second judge pass, order swapped — scores are the average of both: Model B makes all three lines exceptionally crisp and readable, with the requested strong title hierarchy and subtle grain. Model A is also clean and well composed, but its th | Anthropic: Claude Fable 5.1: Both Model A and Model B render all three lines…
Neon night ramen lane
Cinematic anime-inspired street scene at neon night: a cramped midnight ramen stall called 'Blue Kettle No. 9' under rain-slick awnings, with a steaming bowl of squid-ink noodles and three skewers of blistered shishito peppers on the counter as the central subject; magenta pharmacy signage, cyan vending-machine glow, and a flickering amber paper lantern cast mixed colored light across wet asphalt and drifting steam, creating a moody, electric atmosphere; low eye-level composition looking down the narrow lane, crisp details in puddles, 16:9.
OpenAI: GPT-6 Luna: Model A more directly matches the prompt, with clearly visible black squid-ink noodles and three pepper skewers foregrounded beneath the correctly rendered stall name, while retaining the wet neon lane and amber lantern. Model B has a lively, attractive street scene, but its food is less prominent and the bowl does not read as clearly as squid-ink ramen. (Second judge pass, order swapped — scores are the average of both: Model A presents the steaming squid-ink noodles and three pepper skewers mo | Anthropic: Claude Fable 5.1: Model A nails the anime-inspired style, places…
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 isometric perspective and places the bed, centered round rug, desk beneath the back-wall window, and lamp in the front-right area. Model B is more colorful and polished, but uses a frontal perspective rather than the requested isometric view. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested isometric perspective while placing the bed, centered rug, desk beneath the back-wall window, and lamp on the right. Model | Anthropic: Claude Fable 5.1: Model A delivers a true isometric room with the…
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 uses the requested campsite motifs and a restrained palette, but the fire’s glow introduces a clear gradient and the scene feels sparse. Model B is a more complete, balanced campsite illustration with stronger visual depth while staying close to the four-color palette and flat-vector style. (Second judge pass, order swapped — scores are the average of both: Model B delivers a richly balanced campsite scene in a crisp flat-vector style and closely follows the four-color restriction. Model | Anthropic: Claude Fable 5.1: Model A keeps a sparse four-color look but brea…
Glasshouse soda reflections
Hyperreal still life inside a Victorian glasshouse café, centered on a tall ribbed glass of cloudy yuzu soda with a striped paper straw, beside a chrome dessert spoon and a small plate holding two translucent candied violets; every surface must behave correctly: the curved glass refracts the checkerboard tile floor and bends the mint sprig inside the drink, condensation beads distort highlights, the chrome spoon mirrors the greenhouse rafters and nearby apricot macarons, and the surrounding windowpanes show layered reflections of hanging ferns and distant rain; cool overcast daylight, meticulous realism, 16:9.
OpenAI: GPT-6 Luna: Model A more completely realizes the centered still life, including the ribbed, condensation-covered glass, two translucent violets, checkerboard refraction, and convincing spoon and window reflections. Model B includes many of the requested elements, but the scene feels busier and less carefully composed, and the violets and ribbing are less clearly rendered. (Second judge pass, order swapped — scores are the average of both: Model A more clearly delivers the ribbed, condensation-covered soda, | Anthropic: Claude Fable 5.1: Model A delivers nearly every requested element—…
Bakery alley vanishing point
Pastel storybook illustration of a cobbled bakery alley in the port town of Bellmire, viewed from waist height with a strong one-point perspective leading to a tiny clock tower at the far end; in the foreground, a bread cart piled with twelve oversized saffron loaves, a stack of tiny espresso cups, and a child carrying a box labeled 'Morrow Finch Custards' must all be scaled believably against the shopfronts, bicycles, awnings, and receding window rows; keep all architectural lines consistent to a single vanishing point so the alley feels solid and unwarped, with soft morning light and gentle shadows, 16:9.
OpenAI: GPT-6 Luna: Model B better captures the pastel storybook feel and establishes Bellmire as a port town, while also including the bakery signage, bicycles, awnings, and a clear clock-tower vanishing point. Model A has especially convincing perspective and appears closer to the requested loaf count, but Model B is more evocative and visually cohesive overall. (Second judge pass, order swapped — scores are the average of both: Model A gives the alley a clearer, more consistent one-point perspective and includes | Anthropic: Claude Fable 5.1: Model B delivers a more charming pastel storybo…
Macro kumquat tart
Ultra-detailed editorial food photograph, macro close-up of a single kumquat tart on a sea-green crackle-glaze plate, shot at table level with a true macro lens so the sugared kumquat slices, tiny thyme leaves, and pinprick bubbles in the lacquered glaze are razor-sharp while the background falls into creamy blur; a silver dessert fork enters from the lower right with only its front tine in focus, a folded ink-blue napkin and a handwritten receipt reading 'Pier Lantern Bakery, slip 47B' dissolve into shallow bokeh behind, realistic studio color, intimate composition, 16:9.
OpenAI: GPT-6 Luna: Model A better captures the intimate macro close-up, razor-sharp sugared kumquats, crackle-glaze plate, and creamy background blur. Model B renders the receipt text more clearly, but its wider, more documentary framing departs from the requested macro look. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested 16:9 composition and renders the bakery receipt clearly, while keeping the tart, thyme, and crackle-glaze plate convincingly detailed. Model A h | Anthropic: Claude Fable 5.1: Model A delivers the true macro, table-level in…
Matchup powered by OpenRouter.