Head to head: Bytedance Seedream V4.5 Edit vs Google: Nano Banana Pro (Gemini 3 Pro Image)
Bytedance Seedream V4.5 Edit vs Google: Nano Banana Pro (Gemini 3 Pro Image)
By Ryan Merket · Published
A prompt-following face-off across image editing, typography, exact counts, anatomy, and cinematic scenes. The key question is whether polished atmosphere can compete with reliably landing the requested details and composition.
Google: Nano Banana Pro (Gemini 3 Pro Image) takes the saved verdict, **66.1 to 57.2**, with seven task wins to Seedream’s one and no ties. The result is decisive on this task sample, though it should not be read as a claim about image models in general. Nano Banana Pro’s edge is most apparent where the prompt leaves little room for approximation. It won both text tasks, the exact-counting test, and the locksmith bench scene, with stronger legibility, framing, or adherence to the requested details. It also prevailed on reflections and glass, where the evaluations noted a cleaner arrangement and visible reflections, and on the tram-stop perspective, where the requested layout was more faithfully rendered. Seedream’s one win came in hands and anatomy: its close-up made the bracelet work and threads stand out, even though the crop obscured part of the grip. Its cinematic instincts showed again in the courier leap, but Nano Banana Pro’s image made the bunny-hop, lifted sneaker, swinging bag, and surrounding action easier to read—and won that task. **Final call: Nano Banana Pro wins this saved task sample on consistent prompt execution, especially for text, counting, and specified scene layouts. Seedream can deliver striking images, but here its visual drama did not make up for missing or less clearly presented requirements.**
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 emphasizes the requested reflective close-up, with the window and red apple visible in the chrome and reflections around the water glass. Model B is cleaner and more natural in its arrangement, but the specified reflections are less evident on the water’s surface. (Second judge pass, order swapped — scores are the average of both: Model B clearly shows the chrome teapot, water glass, marble counter, window, and an actual red apple, with strong window and apple reflections in | Anthropic: Claude Fable 5.1: Model A shows the window and apple clearly in b…
Macro Locksmith Bench
An extreme macro close-up photo of an urban locksmith’s workbench inside a narrow street-side kiosk, shot with a true macro lens so only the teeth of a freshly cut brass key stamped with tiny serial marks are razor sharp while blurred racks of duplicate keys and a neon OPEN sign melt into creamy bokeh behind it; tactile metal shavings, a red felt mat, and a small dab of machine oil visible, dramatic sidelight from the kiosk window, shallow depth of field, intimate realistic photography
OpenAI: GPT-6 Luna: Model A more fully captures the kiosk workbench details, including the red felt, metal shavings, oil, duplicate-key racks, and glowing OPEN sign, with strong macro texture and atmospheric sidelight. Model B has a clear close-up subject, but the sign is reversed and the scene omits the visible dab of oil, while its key is in focus across more than just the teeth. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the requested macro workbench scene, w | Anthropic: Claude Fable 5.1: Model B delivers a more convincing macro photog…
Night Market Poster Text
A realistic nighttime photo of a poster pasted on a corrugated metal shutter at a street market entrance, lit by a single overhead bulb and framed straight-on so the typography is perfectly readable; the poster design is bold but clean, with legible text that must appear exactly as: "ORBIT NOODLE LAB" on the top line, "MIDNIGHT BOWLS" below it, and a small footer line reading "7 Plover Lane"; high contrast ink colors, slight paper wrinkles, urban grit around the edges, no extra words or misspellings
OpenAI: GPT-6 Luna: Model B presents the required wording more cleanly and legibly on a centered poster, with a convincing nighttime market setting and single overhead bulb. Model A is also atmospheric, but its oversized, tightly stacked typography makes the poster less clean and the text harder to parse at a glance. (Second judge pass, order swapped — scores are the average of both: Model B better captures the gritty, lively market entrance under a single bare bulb, while keeping all requested wording readable. Mo | Anthropic: Claude Fable 5.1: Both render every line of text exactly and legi…
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 seven espresso-filled cups in a straight row, but the cups are oversized and partly cropped at the edges. Model B presents exactly seven separate, clearly countable cups with cleaner spacing and framing, though they are empty. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly seven identical, clearly separated cups in a straight overhead row on white marble. Model A has a polished look, but the cups are oversized, crowded along the bottom edg | Anthropic: Claude Fable 5.1: Both Model A and Model B render exactly seven i…
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 B renders the requested wording clearly in the intended three-line hierarchy on a deep navy background. Model A is legible, but splits the headline across two lines and presents the design as a photographed sign rather than a minimalist poster. (Second judge pass, order swapped — scores are the average of both: Model B renders all three requested lines correctly, crisply, and centered on a deep navy background. Model A has an attractive sign presentation, but breaks the headline across two | Anthropic: Claude Fable 5.1: Model B delivers exactly what was requested: a…
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 A has an appealing close-up and clearly shows bracelet threads, but the extreme crop and partly obscured fingers make the grip less easy to read. Model B presents both hands and the bracelet in a more natural, plausible working pose, with soft daylight and a candid setting. (Second judge pass, order swapped — scores are the average of both: Model B more fully captures a natural candid bracelet-making moment, with both hands clearly visible in plausible grips and soft daylight in a shallow- | Anthropic: Claude Fable 5.1: Model A shows both hands large, crisply lit, an…
Tram Stop Perspective
A rainy dusk urban street scene in a semi-realistic cinematic digital painting, viewed from curb height at a three-way intersection in Brno, with a bright orange tram approaching from the far center, a tiny lime-green scooter parked in the near right foreground, two pedestrians under a striped awning on the mid-left sidewalk, and a seven-story corner pharmacy receding correctly along the street; strict one-point perspective with all storefronts, crosswalk stripes, overhead wires, and curb lines converging consistently so the tram, scooter, benches, bollards, and people have believable relative scale from foreground to distance, wet reflections glowing under cool blue streetlights and warm shop windows, 16:9
OpenAI: GPT-6 Luna: Model B more closely follows the requested one-point street perspective, with the orange tram centered in the distance, a clearly visible lime scooter in the foreground, and pedestrians beneath the awning. Model A has striking cinematic lighting and reflections, but its tram and scooter placement depart from the prompt and its perspective feels less consistent. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the requested layout, with the orange t | Anthropic: Claude Fable 5.1: Model B delivers a cleaner strict one-point per…
Courier Leap
A vibrant flat-vector illustration of a bike courier mid-action in a dense downtown alley, caught at the exact moment she bunny-hops a collapsed stack of blue delivery crates while one sneaker lifts off the pedal and her mustard messenger bag swings outward; loose receipts and rain droplets arc through the air, wheels slightly motion-blurred, jacket flaring, startled pigeons scattering from a puddle, strong diagonal composition and crisp dynamic energy under bright post-storm sunlight between tall mural-covered brick walls, 16:9
OpenAI: GPT-6 Luna: Model A is exceptionally vivid and cinematic, but Model B more clearly depicts a woman bunny-hopping over collapsed blue crates with her mustard bag swinging and one sneaker off the pedal. Model B also includes the alley, flying receipts, rain, puddle, and scattering pigeons in a legible, energetic composition. (Second judge pass, order swapped — scores are the average of both: Model B more clearly captures the female courier’s bunny-hop, lifted sneaker, swinging mustard bag, collapsed blue crat | Anthropic: Claude Fable 5.1: Model A delivers a far more polished, vibrant f…
Matchup powered by OpenRouter.