Head to head: Google: Nano Banana Pro (Gemini 3 Pro Image) vs Hidream I1 Full
Google: Nano Banana Pro (Gemini 3 Pro Image) vs Hidream I1 Full
By Ryan Merket · Published
A prompt-following showdown across eight image-generation tasks, from emotional close-ups and reflective chrome to exact counts and spatial layouts. The results turn on whether a model can reliably deliver the requested details—not just a compelling visual style.
Nano Banana Pro wins decisively: it scored 67.2 to Hidream I1 Full’s 48.9, with 100% statistical confidence. It took seven task wins, lost none, and tied once. That record reflects a consistent edge in matching the brief, not a single standout image carrying the matchup. The clearest gaps came when prompts demanded several precise elements at once. Nano Banana Pro showed all seven correctly colored gadgets, while Hidream omitted the stylus, key fob, and speaker and added an extra object. In the action scene, Google delivered the electric unicycle, dramatic lean, airborne parcel, and rain-slick market; Hidream substituted a bicycle. Its industrial corridor likewise preserved the requested scale progression and technician where Hidream replaced key scene elements with oversized drones. Google also handled nuanced expression and exact presentation better. Its close-up conveyed wet-eyed relief tipping into a trembling smile, while Hidream’s read more as alarm. Both models produced seven cups, but Google’s row was more uniform; and on the typography task it kept the exact three-line layout, while Hidream broke the headline across two lines and added an underline. Hidream can make polished, appealing images, but polish did not make up for those prompt departures. Spatial layout was the lone tie: Hidream placed the bed against the left wall more clearly, while Google’s image was judged cleaner and more consistent overall. That exception matters, but it doesn’t change the broader result. **Verdict: Nano Banana Pro is the clear winner, with stronger fidelity across nearly every kind of constraint tested.**
Emotion & facial expression
An intimate close-up stop-motion clay portrait of a young hardware engineer at a cluttered midnight workbench, realizing her homemade pocket translator has finally powered on after months of failures; her emotion is unmistakable stunned relief tipping into joyful disbelief, shown through wet eyes, lifted brows, a trembling half-laugh smile, and shoulders dropping as tension releases, while tiny status LEDs glow emerald across scattered circuit boards and a soldering iron cools nearby; warm desk-lamp key light against a dim cobalt room, shallow depth of field.
OpenAI: GPT-6 Luna: Model A better captures the requested stunned relief turning into a trembling, joyful smile, with expressive wet eyes and relaxed shoulders in an intimate close-up. Model B has strong stop-motion styling and clear workshop details, but its open-mouthed expression reads more as alarm than joyful disbelief. (Second judge pass, order swapped — scores are the average of both: Model A more clearly conveys the requested wet-eyed stunned relief tipping into a trembling, joyful smile, with the cobalt ro | Anthropic: Claude Fable 5.1: Model A nails the specified emotion—wet glisten…
Attribute & color binding
A crisp studio product tableau in a clean 16:9 advertising style, showing seven distinct gadgets arranged on staggered acrylic blocks against a soft gray backdrop, each with its own exact color and material that must stay perfectly bound to the correct object: a brushed copper smart ring, a translucent cobalt-blue glass USB drive, a matte lime-green silicone game controller, a polished obsidian-black ceramic earbuds case, a satin lavender anodized-aluminum tablet stylus, a frosted amber resin key fob, and a cream-colored woven-fabric mini speaker; softbox lighting, subtle reflections, precise separation, no color bleeding, no swapped materials.
OpenAI: GPT-6 Luna: Model A clearly presents all seven requested gadgets with their colors and materials correctly bound, arranged on acrylic blocks in a polished gray studio tableau. Model B has appealing lighting but omits several requested objects, including the stylus, key fob, and speaker, and adds an extra copper-colored object. (Second judge pass, order swapped — scores are the average of both: Model A clearly presents all seven gadgets with the requested colors and materials on staggered acrylic blocks, in | Anthropic: Claude Fable 5.1: Model A renders all seven gadgets with their cor…
Action & motion
A high-energy stop-motion claymation scene of a courier on an electric unicycle racing through the rain-slick night market of Byte Alley while snatching a tumbling anti-static parcel marked QX-44 from midair, the rider leaning sharply left with coat tails whipping, one hand outstretched, neon vendor signs streaking into motion blur, puddles spraying from the spinning wheel, startled shoppers and dangling cables swept backward by the speed, dynamic Dutch-angle composition, dramatic magenta-and-cyan cyberpunk lighting, 16:9.
OpenAI: GPT-6 Luna: Model A closely matches the electric unicycle, sharp lean, midair QX-44 parcel, rain-slick market, and vivid cyberpunk motion, with a strong Dutch-angle composition. Model B has appealing stop-motion styling and clear parcel text, but substitutes a bicycle for the unicycle and feels less faithful to the requested dramatic composition. (Second judge pass, order swapped — scores are the average of both: Model A captures the electric unicycle, sharp lean, outstretched hand reaching for the clearly | Anthropic: Claude Fable 5.1: Model A delivers a genuine electric unicycle, 'B…
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 B places the bed clearly against the left wall, with the rug centered, desk beneath the back-wall window, and lamp in the front-right corner. Model A is polished and well composed, but its bed appears set against the back wall rather than the left wall. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested layout more clearly, with the bed against the left wall, a centered round rug, and the desk and floor lamp placed by the window on the right. | Anthropic: Claude Fable 5.1: Model A delivers a clean, consistent isometric r…
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 A more clearly shows both the window and the red apple reflected in the chrome, and its close-up framing and dramatic light suit the prompt well. Model B is polished and well balanced, but the required reflections—especially in the water—are less distinct, and it has a visible artifact along the top edge. (Second judge pass, order swapped — scores are the average of both: Model A more clearly shows the window and red apple reflected in the chrome, while Model B’s required reflections in th | Anthropic: Claude Fable 5.1: Model A delivers a convincingly photoreal chrom…
Perspective & scale
A cinematic 16:9 ultra-detailed illustration of the interior of a startup robotics repair hangar called Kestrel Node-7, viewed from just above floor level near a yellow safety stripe, with a long corridor of work bays receding to a single clear vanishing point at the far loading door; in the foreground, a palm-sized teal diagnostic drone sits beside a knee-high silver tool cart, midground a waist-high orange battery cabinet and a human technician on a rolling stool, farther back a compact delivery robot, then a parked van-sized autonomous forklift, and at the distant end a towering six-meter matte-white assembly mech under scaffolding, all with correct relative scale, straight architectural lines, and believable depth so no object looks warped or wrongly sized; hard industrial LED lighting with cool reflections on polished concrete.
OpenAI: GPT-6 Luna: Model A includes the requested sequence of differently scaled equipment, a technician, a distant assembly mech, and a strong corridor vanishing point, with convincing industrial depth. Model B has an attractive symmetrical hangar but replaces the key scene elements with oversized foreground drones and omits the technician, forklift, and mech. (Second judge pass, order swapped — scores are the average of both: Model A includes the requested sequence of objects at convincing relative scales, with | Anthropic: Claude Fable 5.1: Model A delivers nearly every element of the dep…
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 shows exactly seven clearly separate, identical cups in a neat straight row on white marble. Model B also has seven cups, but the inconsistent, confusing handle shapes make the arrangement look less uniform and polished. (Second judge pass, order swapped — scores are the average of both: Both images show exactly seven cups in a straight row on white marble, but Model A has more consistent spacing and cleaner, more uniform cups. Model B shows espresso in the cups, while Model A’s cups are | Anthropic: Claude Fable 5.1: Model A shows exactly seven identical empty esp…
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 matches the requested three centered lines with crisp, correctly spelled text and a deep navy grainy background. Model B renders the wording legibly but breaks the headline across two lines, adds an underline, and uses a heavier, rougher treatment that departs from the minimalist layout. (Second judge pass, order swapped — scores are the average of both: Model A follows the requested three-line centered layout with crisp, correctly spelled text and a clean deep navy background. Model B i | Anthropic: Claude Fable 5.1: Model A renders all three lines correctly spell…
Matchup powered by OpenRouter.