Head to head: Krea 2 Large vs Lumina Image 2
Krea 2 Large vs Lumina Image 2
By Ryan Merket · Published
An automated head-to-head image matchup. Krea 2 Large scored 61.9, Lumina Image 2 scored 43.0.
### Legible multi-line text Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A presents all three requested phrases clearly with strong hierarchy and centering, though it adds unwanted punctuation and prominent bokeh rather than subtle grain. Model B has major spelling and text-rendering errors, merges the lower content into one illegible line, and fails the requested three-line structure. (Second judge pass, order swapped — scores are the average of both: Model B misspells “MARKET,” severely garbles the remaining copy, and fails the requested three-line structure. | Anthropic: Claude Fable 5.1: Model A renders all three lines correctly… ### Hands & anatomy Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A clearly depicts both hands with convincing anatomy, a plausible thread-tying action, attractive daylight, and strong detail, though it misses the requested 16:9 format. Model B uses the correct aspect ratio and shallow depth of field, but the fingers appear fused and distorted, the grip is unclear, and the image is excessively soft. (Second judge pass, order swapped — scores are the average of both: Model A shows two clearly visible, anatomically plausible hands making a convincing grip | Anthropic: Claude Fable 5.1: Model A delivers a convincing candid shot w… ### Copper Kettle Texture Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A captures substantially more of the requested micro-textures and specific props, including laminated buns, coarse sugar, papery pistachio shells, zested citrus, herringbone linen, and a heavily nicked board, though its square framing and cropped kettle weaken the composition. Model B better matches the 16:9 format and window-lit staging, but substitutes a teapot, generic rolls, bare pistachio kernels, striped cloth, and an unzested citrus half. (Second judge pass, order swapped — scores a | Anthropic: Claude Fable 5.1: Model A nails nearly every prompt element—… ### Bittersweet Victory Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A strongly captures the close claymation portrait, chipped demitasse, mustard apron, and nuanced bittersweet expression through damp eyelids, raised inner brows, flushed cheeks, and an exhausted smile. Model B has polished lighting and composition but reads as contented or smug rather than tearfully relieved, is framed too widely, and renders the menu as distracting gibberish. (Second judge pass, order swapped — scores are the average of both: Model A more convincingly captures the handcra | Anthropic: Claude Fable 5.1: Model A nails the specified micro-expressi… ### Rainy Conservatory Toast Winner: **Tie**. OpenAI: GPT-5.6 Sol Pro: Model A more faithfully includes the sparkling flute, saffron jelly, authentic samovar, rain-specked window, candles, and layered optical effects, though its square framing and awkward spoon placement weaken the spread. Model B has a cleaner 16:9 composition and excellent reflections, but substitutes a teapot for the samovar and liquid for jelly, omitting key prompt details. (Second judge pass, order swapped — scores are the average of both: Model B delivers the stronger cinematic 16:9 magazine | Anthropic: Claude Fable 5.1: Model A nails the specific props (true samo… ### Reflections & glass Winner: **Lumina Image 2**. OpenAI: GPT-5.6 Sol Pro: Model B better matches the requested 16:9 framing and presents the teapot, water glass, window, and apple clearly in a polished composition, with convincing counter reflections. Model A is photorealistic but uses a square frame, crops the apple, and neither model clearly renders both the window and apple in the water’s surface, though Model B comes closer overall. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested 16:9 format, dramatic lighti | Anthropic: Claude Fable 5.1: Model A delivers convincing photorealism a… ### Exact counting Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A correctly shows exactly seven separate espresso cups in one straight row, though the outer cups are cropped, the lighting casts heavy shadows, and the image is not 16:9. Model B has a polished 16:9 presentation but shows ten cups arranged in two rows, fundamentally failing the exact-count and layout requirements. (Second judge pass, order swapped — scores are the average of both: Model A shows exactly seven identical, separately countable cups in one straight row, though the outer cups a | Anthropic: Claude Fable 5.1: Model A shows seven identical espresso cup… ### Midnight Dumpling Assembly Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A clearly depicts six distinct cooks performing nearly all specified simultaneous actions, with convincing stop-motion materials, steam, harbor lighting, and richly readable props, though it misses the requested 16:9 format. Model B uses the correct widescreen composition but omits or weakens several key actions and nighttime harbor details, and its smooth illustration style is less convincingly handcrafted. (Second judge pass, order swapped — scores are the average of both: Model A captur | Anthropic: Claude Fable 5.1: Model A nails nearly every prompt element… ### Legible multi-line text Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A delivers all three centered, highly legible text lines with correct wording, strong hierarchy, and an appropriately dark navy treatment, though the texture is heavier than subtle and the separator resembles a bullet. Model B contains extensive misspellings and malformed text, splits the title incorrectly, and fails the core legibility requirement despite a broadly suitable background. (Second judge pass, order swapped — scores are the average of both: Model A renders all three lines corr | Anthropic: Claude Fable 5.1: Model A renders all three lines perfectly… ### Hands & anatomy Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A clearly depicts a colorful friendship bracelet with two anatomically plausible hands and a convincing grip, though it misses the requested 16:9 framing. Model B matches the aspect ratio and has attractive shallow focus, but the plain cord and simple knot do not read strongly as a friendship bracelet. (Second judge pass, order swapped — scores are the average of both: Model A clearly depicts a colorful friendship bracelet with two anatomically plausible hands and a convincing grip, though | Anthropic: Claude Fable 5.1: Model B delivers the requested 16:9 frame,… ### Copper Kettle Texture Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A captures the hammered copper, herringbone linen, pistachio shells, knife-scarred board, laminated buns, coarse sugar, and grazing cool light with notably strong micro-texture, though it misses the requested 16:9 framing and is somewhat cluttered. Model B has cleaner widescreen composition, but its polished teapot, warm lighting, unshelled cardamom-like seeds, unzested citrus, and plain rolls diverge substantially from the specified workstation. (Second judge pass, order swapped — scores | Anthropic: Claude Fable 5.1: Model A delivers a hammered dented copper k… ### Bittersweet Victory Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A closely captures the requested bittersweet emotional cues—including damp lower eyelids, raised inner brows, flushed cheeks, and an exhausted smile—while also clearly presenting the chipped demitasse, mustard apron, blurred menu, and mixed warm/cool lighting. Model B is polished and well composed, but its broad cheerful expression lacks the specified tearful relief and exhaustion, and its style appears more like smooth CGI than tactile claymation. (Second judge pass, order swapped — score | Anthropic: Claude Fable 5.1: Model A nails the nuanced emotional brief… ### Rainy Conservatory Toast Winner: **Tie**. OpenAI: GPT-5.6 Sol Pro: Model A more faithfully depicts the flute, visibly set saffron jelly, true samovar, glass-roofed conservatory, rain, and layered optical effects, though it misses the requested 16:9 format and feels tightly cropped. Model B has a cleaner 16:9 magazine composition, but the flute resembles a tumbler, the coupe appears liquid-filled, and the glass roof is absent. (Second judge pass, order swapped — scores are the average of both: Model B delivers the stronger refined magazine composition, correct 1 | Anthropic: Claude Fable 5.1: Model A delivers a convincing samovar, vis… ### Reflections & glass Winner: **Lumina Image 2**. OpenAI: GPT-5.6 Sol Pro: Model B better satisfies the requested 16:9 framing and offers a more balanced, dramatic composition with strong photorealistic polish. Model A more clearly shows the apple and window in the chrome, but neither model convincingly reflects both elements on the water’s surface, and Model A uses a square aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B better satisfies the requested 16:9 framing and delivers a more balanced, dramatic composition with strong | Anthropic: Claude Fable 5.1: Model A clearly renders the red apple and w… ### Exact counting Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A shows exactly seven cups in a straight row, but the end cups are cropped, the cups touch, and the image is not 16:9. Model B has cleaner spacing and the requested aspect ratio, but it contains only five cups, a decisive failure for an exact-counting prompt. (Second judge pass, order swapped — scores are the average of both: Model B is clean, attractive, and correctly framed at 16:9, but it contains only five cups rather than exactly seven. Model A contains seven cups in a straight row, b | Anthropic: Claude Fable 5.1: Model A does contain seven identical espre… ### Midnight Dumpling Assembly Winner: **Krea 2 Large**. OpenAI: GPT-5.6 Sol Pro: Model A clearly depicts six cooks performing nearly all the specified simultaneous tasks, with strong stop-motion character, readable staging, harbor lighting, and detailed ingredients, though it misses the requested 16:9 format and the sneaked bite is not clearly pickled mango. Model B uses the correct widescreen format and a balanced kitchen layout, but shows only five cooks, omits several key actions, and has a conspicuously headless central figure. (Second judge pass, order swapped — scores | Anthropic: Claude Fable 5.1: Model A delivers all six distinct actions (…
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-5.6 Sol Pro: Model A presents all three requested phrases clearly with strong hierarchy and centering, though it adds unwanted punctuation and prominent bokeh rather than subtle grain. Model B has major spelling and text-rendering errors, merges the lower content into one illegible line, and fails the requested three-line structure. (Second judge pass, order swapped — scores are the average of both: Model B misspells “MARKET,” severely garbles the remaining copy, and fails the requested three-line structure. | Anthropic: Claude Fable 5.1: Model A renders all three lines correctly…
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-5.6 Sol Pro: Model A clearly depicts both hands with convincing anatomy, a plausible thread-tying action, attractive daylight, and strong detail, though it misses the requested 16:9 format. Model B uses the correct aspect ratio and shallow depth of field, but the fingers appear fused and distorted, the grip is unclear, and the image is excessively soft. (Second judge pass, order swapped — scores are the average of both: Model A shows two clearly visible, anatomically plausible hands making a convincing grip | Anthropic: Claude Fable 5.1: Model A delivers a convincing candid shot w…
Copper Kettle Texture
A hyper-detailed editorial food photograph of a pastry chef’s workstation after dawn service: a dented copper sugar kettle with hammered dimples, a folded indigo herringbone linen towel, scattered pistachio shells with papery interiors, a half-zested bergamot, a worn walnut cutting board with knife nicks, and three cardamom buns showing laminated layers and coarse sugar crystals; side-lit by cool winter window light with a narrow sunbeam grazing the surfaces, shallow camera angle across the table, crisp focus on micro-textures, 16:9.
OpenAI: GPT-5.6 Sol Pro: Model A captures substantially more of the requested micro-textures and specific props, including laminated buns, coarse sugar, papery pistachio shells, zested citrus, herringbone linen, and a heavily nicked board, though its square framing and cropped kettle weaken the composition. Model B better matches the 16:9 format and window-lit staging, but substitutes a teapot, generic rolls, bare pistachio kernels, striped cloth, and an unzested citrus half. (Second judge pass, order swapped — scores a | Anthropic: Claude Fable 5.1: Model A nails nearly every prompt element—…
Bittersweet Victory
A stylized stop-motion claymation close portrait of a young barista in a mustard apron holding a chipped ceramic demitasse at chest height, conveying unmistakable relieved pride after finally perfecting a difficult roast profile; the expression should show damp lower eyelids, a tiny exhausted smile, raised inner brows, flushed cheeks, and a loosened jaw, with a hand-lettered café menu blurred behind them, warm tungsten pendant light from above and soft rainy daylight from a side window shaping the face.
OpenAI: GPT-5.6 Sol Pro: Model A strongly captures the close claymation portrait, chipped demitasse, mustard apron, and nuanced bittersweet expression through damp eyelids, raised inner brows, flushed cheeks, and an exhausted smile. Model B has polished lighting and composition but reads as contented or smug rather than tearfully relieved, is framed too widely, and renders the menu as distracting gibberish. (Second judge pass, order swapped — scores are the average of both: Model A more convincingly captures the handcra | Anthropic: Claude Fable 5.1: Model A nails the specified micro-expressi…
Rainy Conservatory Toast
An elegant cinematic still life in a glass-roofed tea conservatory during a storm: a tall flute of sparkling quince soda, a crystal coupe filled with saffron jelly, a polished chrome samovar, and a rain-specked window revealing blurred topiary outside; the transparent glass, liquid meniscus, internal bubbles, mirrored metal, and wet reflections must all behave optically correctly, with lightning-muted blue daylight and amber table candles creating layered refractions and highlights, composed as a refined magazine spread, 16:9.
OpenAI: GPT-5.6 Sol Pro: Model A more faithfully includes the sparkling flute, saffron jelly, authentic samovar, rain-specked window, candles, and layered optical effects, though its square framing and awkward spoon placement weaken the spread. Model B has a cleaner 16:9 composition and excellent reflections, but substitutes a teapot for the samovar and liquid for jelly, omitting key prompt details. (Second judge pass, order swapped — scores are the average of both: Model B delivers the stronger cinematic 16:9 magazine | Anthropic: Claude Fable 5.1: Model A nails the specific props (true samo…
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-5.6 Sol Pro: Model B better matches the requested 16:9 framing and presents the teapot, water glass, window, and apple clearly in a polished composition, with convincing counter reflections. Model A is photorealistic but uses a square frame, crops the apple, and neither model clearly renders both the window and apple in the water’s surface, though Model B comes closer overall. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested 16:9 format, dramatic lighti | Anthropic: Claude Fable 5.1: Model A delivers convincing photorealism a…
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-5.6 Sol Pro: Model A correctly shows exactly seven separate espresso cups in one straight row, though the outer cups are cropped, the lighting casts heavy shadows, and the image is not 16:9. Model B has a polished 16:9 presentation but shows ten cups arranged in two rows, fundamentally failing the exact-count and layout requirements. (Second judge pass, order swapped — scores are the average of both: Model A shows exactly seven identical, separately countable cups in one straight row, though the outer cups a | Anthropic: Claude Fable 5.1: Model A shows seven identical espresso cup…
Midnight Dumpling Assembly
A lively overhead-to-three-quarter 16:9 scene in handcrafted stop-motion style showing six night-shift cooks around a stainless prep table in a tiny harbor canteen, all interacting at once: one pinching green dough around squid filling, one laughing while flouring a rolling pin, one passing a tray of crescent dumplings, one checking a steaming bamboo basket, one scribbling orders on pink tickets, and one sneaking a bite of pickled mango; balance the group naturally with bowls of black vinegar, chopped chives, striped aprons, stacked enamel plates, and a porthole window with sodium-vapor dock light, creating a coherent, busy but readable composition.
OpenAI: GPT-5.6 Sol Pro: Model A clearly depicts six distinct cooks performing nearly all specified simultaneous actions, with convincing stop-motion materials, steam, harbor lighting, and richly readable props, though it misses the requested 16:9 format. Model B uses the correct widescreen composition but omits or weakens several key actions and nighttime harbor details, and its smooth illustration style is less convincingly handcrafted. (Second judge pass, order swapped — scores are the average of both: Model A captur | Anthropic: Claude Fable 5.1: Model A nails nearly every prompt element…
Matchup powered by OpenRouter.