Head to head: FLUX.2 [klein] 9B vs Fibo Lite

This one was decided on prompt discipline, not vibes. Fibo Lite takes a narrow but real edge by converting more of the fussy, failure-prone instructions into correct images, while FLUX.2 [klein] 9B remains the cleaner stylist in a couple of standout scenes.

The scoreline says what the images do: this was close, but not random. Fibo Lite finishes ahead **66.6 to 64.4**, wins **4 tasks to 2**, with **2 ties**, and the aggregate verdict lands at **77% confidence**—a lean, not a rout. That matters here, because neither model dominated the set; Fibo Lite simply made fewer prompt-breaking mistakes in the tasks where adherence mattered most. Its best work came in the precision tests. In **Reflections & glass**, Fibo Lite was the one that actually delivered the required reflected window and apple interplay in the chrome teapot and glass/water area, where FLUX looked polished but less explicit. In **Attribute binding**, the difference was even simpler: Fibo Lite placed the single yellow duck, green cube, red sphere, and blue cylinder correctly, while FLUX invented an extra duck and broke the prompt. **Negation** and **Nine Mismatched Teacups** followed the same pattern—Fibo Lite wasn’t perfect, but it was closer to the assignment, and closer counts. FLUX.2 [klein] 9B, though, was not outclassed. It won **Perspective & scale** by being more convincingly photorealistic, with cleaner one-point geometry and better respect for the requested framing. It also took **Rolling Colander Spill** decisively, because it captured the editorial-action brief—the mid-roll colander, kumquats, towel motion, droplets, and diagonal energy—while Fibo Lite drifted into the wrong fruit and a flatter sense of motion. If you care most about compositional polish and cinematic control, FLUX still has a case. The two ties explain why this matchup stays in "lean" territory. In **Rainy Workbench Realism** and **Midnight Repair Table**, the judges split on what mattered more: mood versus prompt specifics, noir atmosphere versus staging accuracy, workshop authenticity versus framing discipline. That is usually the sign of two models trading different strengths rather than one clearly separating. **Final call: Fibo Lite wins on instruction fidelity. FLUX.2 [klein] 9B is often prettier and occasionally sharper, but Fibo Lite was more reliable at getting the actual brief right—and in a head-to-head this close, that is the difference.**

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.

FLUX.2 [klein] 9B:

Fibo Lite:

Model B better satisfies the core reflection requirement: the window is convincingly reflected in the chrome teapot and the red apple is visibly echoed in both the teapot and the glass/water area, while Model A has a strong look but weaker or less explicit apple reflection in the water. Model B also has a more balanced close-up composition and richer photorealistic lighting, though Model A is cleaner and more polished overall. (Second judge pass, order swapped — scores are the average of both: Model B better satisfies the prompt by clearly showing the window and the red apple reflected in the chrome teapot and also visible through/reflected in the glass of water, while maintaining a convincing polished marble surface and dramatic side light. Model A is attractive and cleanly composed, but the required reflections are less explicit and less accurate—especially the apple/window interplay in the water and chrome—so it adheres less strongly to the prompt.)

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.

FLUX.2 [klein] 9B:

Fibo Lite:

Model B follows the prompt more faithfully: it has one yellow duck on top of the matte green cube, the glossy red sphere to the right, and the small blue cylinder behind the main objects. Model A is visually polished, but it violates the prompt by adding a second duck on the red sphere, reducing adherence despite strong composition and rendering. (Second judge pass, order swapped — scores are the average of both: Model B follows the prompt more accurately: the matte green cube is left of the glossy red sphere, the blue cylinder sits behind them, and there is a single yellow rubber duck on top of the cube. Model A is clean and attractive, but it violates the prompt by showing two ducks and placing one on the red sphere instead of having a single duck on the cube.)

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.

FLUX.2 [klein] 9B:

Fibo Lite:

Model A better matches the requested photorealistic one-point perspective, with very consistent shelf recession, book scaling, and warm overhead lighting in a clear 16:9 frame. Model B is also strong and attractive, but it feels slightly more stylized than photorealistic and does not match the requested 16:9 aspect ratio as well. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested photorealistic one-point perspective, with shelves and lights converging cleanly to a single vanishing point and more convincing real-world scale throughout the aisle. Model B is strong and attractive, but it feels slightly more stylized and its geometry and lighting are a bit less natural than Model A's.)

Negation

A cozy reading nook with an armchair, a stack of books, and a mug of tea by a window — with absolutely NO plants, NO lamps, and NO artwork or picture frames anywhere in the frame. Warm afternoon light, 16:9.

FLUX.2 [klein] 9B:

Fibo Lite:

Model B adheres slightly better to the prompt by clearly presenting a cozy reading nook with an armchair, books, and a mug by the window while avoiding plants, lamps, and artwork. Model A is also strong and visually polished, but the mug perched on the book stack feels a bit less natural for the requested nook setup and the overall composition is slightly less intimate. (Second judge pass, order swapped — scores are the average of both: Model B adheres more closely to the prompt, showing a cozy armchair, books, and tea by a window with warm afternoon light while avoiding plants, lamps, and artwork. Model A is attractive and well composed, but it introduces an extra side table not mentioned in the prompt and includes visible text on book spines, making it feel slightly less clean and precise in adherence.)

Nine Mismatched Teacups

Top-down still-life photograph on a faded sea-green tabletop showing EXACTLY nine mismatched vintage teacups, each fully visible and individually countable, arranged in an irregular spiral with enough spacing to avoid overlap; every cup has a different pattern and handle shape, two contain amber tea, three contain loose buttons, one holds a tiny lemon, and the remaining cups are empty; soft overcast daylight, realistic textures, clean composition

FLUX.2 [klein] 9B:

Fibo Lite:

Model A has a pleasing top-down still life and the sea-green tabletop, but it shows only eight teacups and misses the requested irregular spiral arrangement. Model B gets the exact count of nine and clearer individual visibility with varied cups, though it forms a ring rather than a spiral and misplaces the lemon in a tea-filled cup while also underdelivering on the specified button/tea distribution. (Second judge pass, order swapped — scores are the average of both: Model B is closer to the requested top-down still life with vintage mismatched teacups on a faded sea-green tabletop, and it presents a cleaner, more cohesive arrangement, though it misses the exact object counts and puts the lemon in a tea-filled cup. Model A has appealing spacing and realism, but it clearly fails the requirement of exactly nine cups and the specified distribution of tea, buttons, lemon, and empty cups, while also not forming the requested irregular spiral.)

Rolling Colander Spill

A stainless-steel colander has just bounced off a narrow kitchen counter and is captured mid-roll in the air, flinging rinsed kumquats, basil leaves, and sparkling droplets outward in different arcs above a checkerboard linoleum floor, while a striped dish towel whips off the edge behind it; high-speed editorial food photography, dramatic side lighting, crisp frozen motion, low camera angle, dynamic diagonal composition, 16:9

FLUX.2 [klein] 9B:

Fibo Lite:

Model A matches the prompt much more closely with a stainless-steel colander captured mid-roll, orange kumquats, basil leaves, sparkling droplets, a striped towel whipping behind, checkerboard floor, and strong dynamic diagonal energy. Model B has a pleasing look but misses key prompt details: the fruit reads more like limes than kumquats, the towel is not whipping off the edge, the composition is less dynamic, and the scene does not clearly show the colander bouncing off a narrow counter in a high-speed editorial style. (Second judge pass, order swapped — scores are the average of both: Model A matches the prompt much more closely with a stainless-steel colander captured mid-roll, kumquats, basil, droplets, a checkerboard floor, and a striped towel whipping behind it in a dynamic low-angle editorial composition. Model B is attractive and technically solid, but it misses key prompt details by showing green citrus instead of kumquats, a less convincing rolling action, and a weaker sense of dramatic diagonal motion.)

Rainy Workbench Realism

Photorealistic close interior scene of a rainy afternoon workshop corner: a scratched copper kettle sits on a worn oak workbench beside a folded indigo dishcloth, a half-peeled tangerine, and a carpenter’s hand resting near a steel ruler, with raindrop-blurred window light casting believable soft shadows and cool reflections across metal, wood grain, skin, and fabric; naturalistic color, shallow depth of field, physically accurate materials, cinematic 16:9

FLUX.2 [klein] 9B:

Fibo Lite:

Model B better matches the rainy workshop mood and physical realism, with more convincing raindrop-blurred window light, cooler reflections, and a more cinematic, tactile rendering of the materials. Model A is strong and cleanly composed, but it feels slightly brighter and less workshop-authentic, and the tangerine is less clearly half-peeled than in Model B. (Second judge pass, order swapped — scores are the average of both: Model A adheres more closely to the prompt with a clearer worn oak workbench, more believable carpenter context, and a stronger cinematic 16:9 composition while maintaining realistic rainy window light and material rendering. Model B is moodier and attractive, but it misses the specified aspect ratio and the tangerine appears more peeled than half-peeled, with a slightly less convincing workshop setup.)

Midnight Repair Table

Film-noir black-and-white scene of four subjects gathered around a cramped apartment dining table during a late-night household repair: an elderly woman threading a needle, a teenage boy holding a desk lamp toward a torn lampshade, a one-eyed tabby cat pawing at a fallen spool, and a neighbor in raincoat steadying a wobbling chair; the table is crowded with scissors, teacups, buttons, and folded fabric, and the composition should feel coherent, balanced, and full of subtle interaction, with hard window-blind shadows and a slightly wide cinematic framing, 16:9

FLUX.2 [klein] 9B:

Fibo Lite:

Model B adheres more closely to the prompt by clearly showing the torn lampshade under a desk lamp, the neighbor actively steadying a wobbling chair, stronger noir blind-shadow lighting, and denser table clutter with subtle interaction among all four subjects. Model A is clean and cinematic, but it misses key specifics like the cat pawing at a fallen spool and the neighbor’s chair interaction, and its noir mood is less pronounced. (Second judge pass, order swapped — scores are the average of both: Model A adheres more closely to the prompt by clearly showing the neighbor in a raincoat steadying a wobbling chair, the teenage boy holding the desk lamp toward the torn lampshade, and a wider 16:9 cinematic framing with coherent interaction among all four subjects. Model B is moodier and attractive, but it misses key prompt details such as the one-eyed cat, the chair-steadying action, and the specified wide framing, making Model A the stronger overall match.)

Matchup powered by OpenRouter.