Head to head: AuraFlow vs Krea 2 Large
AuraFlow vs Krea 2 Large
One model brought flashes of style; the other brought repeatable prompt obedience. This matchup wasn’t close in the aggregate, and the task sheet shows exactly why.
AuraFlow’s problem here is simple: it too often optimized for vibe over instructions. Its lone win came on **Perspective & scale**, where it delivered the cleanest one-point library aisle with convincing recession, warm lighting, and the right photoreal framing. That result matters, because it shows AuraFlow can still lock in when spatial discipline is the whole assignment. But across the rest of the board, Krea 2 Large was the more trustworthy image model by a wide margin. It took **7 of 8 tasks**, won the aggregate **68.1 to 46.4**, and the statistical read is blunt: **98% confidence**, decisive. The pattern is consistent rather than flashy. Krea was better at exact compliance on the restricted-palette campsite, the seven spice jars, the exact-count espresso cups, the glassworks catwalk, the no-screens harbor radar room, the legible multi-line text poster, and the Memphis pattern archivist scene. What stands out is the nature of AuraFlow’s misses. They weren’t subtle quality disputes; they were prompt-fidelity failures. It drifted into a badge-like composition instead of a true 16:9 campsite scene, muddled the spice contents, completely blew the exact cup count, inserted a forbidden radar screen, and collapsed on legible text. In other words, AuraFlow often made attractive images that stopped being the requested images. Krea 2 Large wasn’t perfect—there were small misses, like tightly packed cups, tiny off-palette highlights, or a digital-looking desk device in the claymation setup—but those are second-order flaws. On the core question of whether the model can follow concrete instructions, count correctly, preserve scene constraints, and render usable text, Krea was plainly more dependable. **Final call: Krea 2 Large wins decisively. AuraFlow has isolated strengths in perspective-heavy realism, but Krea is the model you pick when prompt adherence actually matters.**
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.
Model B better matches the mountain campsite-at-dusk brief with a stronger 16:9 landscape composition and a more balanced, engaging scene, while staying close to the restricted cream/burnt orange/teal/charcoal palette. Model A is attractive and clean, but it feels more like a centered badge/emblem than a balanced widescreen illustration and appears less compliant with the requested aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested mountain campsite at dusk in a flat-vector 16:9 composition and stays close to the four-color palette, though there are a few tiny off-palette-looking highlights. Model A is attractive and clean, but it is not 16:9, feels more like a badge/poster composition than a balanced scene, and appears to introduce extra tonal variations beyond the strict four-color constraint.)
Seven Sample Jars
A food-lab technician in a spotless test kitchen arranges EXACTLY seven distinct spice sample jars in a neat row on a stainless counter for catalog photography—one jar each of saffron threads, smoked paprika, green cardamom pods, black sesame seeds, dried lavender, pink peppercorns, and ground turmeric—every jar fully visible, separated, and individually countable, shot in soft side light with shallow shadows, clean editorial composition, 16:9.
Model B adheres more closely to the prompt by showing exactly seven distinct, fully visible, individually countable jars with contents that better match the requested spices in a clean editorial setup. Model A is visually striking, but it appears to include mismatched contents and a less accurate set of spices despite strong composition and polish. (Second judge pass, order swapped — scores are the average of both: Model B adheres closely to the prompt with exactly seven fully visible, separated jars in a neat row on a stainless counter, and the contents largely match the requested spices in a clean editorial setup. Model A also shows seven jars, but several contents are incorrect or ambiguous, one spice appears spilled outside its jar, and the composition feels less like catalog photography despite stronger dramatic styling.)
Perspective & scale
A photorealistic one-point-perspective shot down a long empty library aisle, shelves receding to a single vanishing point, books getting consistently smaller with distance and nothing warped or wrongly sized, warm overhead light, 16:9.
Model A matches the prompt much better with a clear one-point perspective down a long empty library aisle, consistent scale recession, warm overhead lighting, and a clean photorealistic look. Model B has strong depth but feels like a cramped archive stack rather than the requested aisle, is darker and less warm, and its composition is more cluttered and less aligned with the prompt. (Second judge pass, order swapped — scores are the average of both: Model A better matches the prompt with a clear photorealistic one-point-perspective library aisle, consistent scale recession, warm overhead lighting, and a cleaner, more balanced composition. Model B has strong depth but feels more like a cramped archive stack than a typical library aisle, with a darker, less polished presentation and a square framing rather than the requested 16:9 feel.)
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.
Model A fails the exact-counting requirement badly, showing many more than seven cups and not a single straight row, though it is visually polished. Model B correctly presents exactly seven identical cups in one straight row on white marble, but it misses the overhead flat-lay view and the cups are touching rather than clearly separate. (Second judge pass, order swapped — scores are the average of both: Model B is much closer to the prompt by showing exactly seven identical espresso cups in a single straight row on white marble, though the view is not strongly overhead and the cups are tightly packed rather than clearly separate. Model A is aesthetically pleasing and more overhead, but it fails the exact counting and arrangement requirements by showing many more than seven cups in a grid instead of one straight row.)
Glassworks Catwalk Perspective
A wide 16:9 view down a blown-glass factory catwalk in one-point perspective, where the railings, overhead trusses, and floor grates all converge cleanly toward a single distant vanishing point; in the foreground a glassblower carrying a long punty rod appears largest, midground assistants rolling molten gathers look proportionally smaller, and far below on the furnace floor tiny workers move between annealing ovens and tool benches, all lit by fiery orange furnace glow mixed with cool skylight, with correct relative scale and no warped architecture.
Model B better matches the requested wide catwalk view with clear one-point perspective, layered worker scaling from foreground to far below, and a convincing mix of furnace glow and skylight. Model A has strong symmetry and mood, but it feels more like a narrow corridor than a broad factory catwalk, the worker scaling/story beats are less accurate, and some architectural elements feel less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested wide catwalk view with clear one-point perspective, layered worker scaling from foreground to furnace floor, and a convincing mix of orange furnace light and cool skylight. Model A has strong symmetry and mood, but it feels more like a narrow corridor than a broad factory catwalk, misses the far-below worker hierarchy, and introduces distracting molten objects on the walkway that weaken prompt fidelity.)
Harbor Radar Room Without Screens
Inside a small harbor radar office at dawn, a marine safety inspector in an orange knit sweater updates paper navigation logs beside rain-speckled windows overlooking tugboats and channel buoys, lit by cool blue morning light and one warm desk lamp, tactile claymation / stop-motion textures with handmade props and slightly imperfect fingerprints in the materials, composed as a quiet over-the-shoulder workplace scene in 16:9, and the room must contain NO digital screens, NO monitors, and NO televisions anywhere.
Model B adheres much better to the claymation stop-motion brief, the over-the-shoulder workplace composition, and the no-screens requirement while preserving the dawn harbor mood with rain-speckled windows and paper logs. Model A is visually polished, but it clearly includes a radar monitor, violating the core constraint, and feels more CG-rendered than handmade claymation. (Second judge pass, order swapped — scores are the average of both: Model B matches the claymation over-the-shoulder harbor office scene well, with rain-speckled windows, paper logs, tugboats and buoys, and strong handmade texture cues; its main prompt miss is the presence of a digital-looking desk device. Model A has pleasing composition and lighting, but it clearly violates the prompt by including a large radar monitor/screen and also departs from the specified tactile stop-motion handmade look.)
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.
Model B closely matches the prompt with three centered lines of crisp, correctly spelled sans-serif text on a deep navy background and pleasing minimal composition. Model A has severe text rendering errors, extra unintended text, and broken hierarchy, so it fails the key requirement of legible multi-line text despite a somewhat fitting color palette. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the prompt with all three lines centered, correctly spelled, and clearly legible on a deep navy background with a clean minimalist feel. Model A has severe text rendering errors, extra unintended text, and a much noisier composition that breaks the requested poster design.)
Memphis Pattern Archivist
A pattern archivist cataloging laminate swatches in a midnight print studio, rendered in unmistakable 1980s Memphis design style with bold squiggles, terrazzo speckles, asymmetrical geometric shapes, pastel-and-neon blocks, black zigzags, and playful postmodern furniture; the worker stands at a tall drafting table under a cone pendant lamp, with pinned sample cards and a rolling flat file behind them, crisp graphic edges, exaggerated color blocking, dynamic diagonal composition, 16:9.
Model B adheres more closely to the prompt by clearly depicting a pattern archivist in a midnight print studio with pinned sample cards, a rolling flat file, a tall drafting table under a cone pendant lamp, and a stronger unmistakable 1980s Memphis vocabulary. Model A has bold Memphis-inspired color blocking and a dynamic look, but it feels more like a photoreal studio scene with weaker archival details and less of the playful postmodern furniture and pinned-swatch setup requested. (Second judge pass, order swapped — scores are the average of both: Model B matches the prompt more completely, with a clear Memphis-style print studio scene featuring laminate swatches, pinned sample cards, a rolling flat file, a tall drafting table, and crisp graphic color blocking in a dynamic composition. Model A has strong Memphis cues and striking visuals, but it feels more like a retro photo-collage than crisp graphic illustration, and it misses or weakens several requested studio details such as the pinned sample cards and rolling flat file.)
Matchup powered by OpenRouter.