Head to head: DeepSeek Janus-Pro vs Fibo Bbq Preview
DeepSeek Janus-Pro vs Fibo Bbq Preview
By Ryan Merket · Published
This matchup tests whether visual polish survives demanding instructions involving spatial layout, negation, anatomy, perspective, attribute binding, and narrative detail. The results expose a substantial gap in prompt control, despite recurring flaws shared by both models.
Fibo Bbq Preview wins decisively: **62.3 to 38.8**, with **limited confidence**. The task ledger is even more emphatic—Fibo takes 15 of 16 verdicts, with one tie and no outright wins for DeepSeek Janus-Pro. The difference is instruction control. Fibo consistently handled the centered controller bench layout, kept the duck bound to the green cube, respected the plant/lamp/artwork negations, produced more convincing bracelet-tying hands, and built cleaner one-point perspective. It also captured far more of the prompt-specific machinery in the courier scene and the intended vector geometry, palette, tram emphasis, kiosks, and antenna sculptures in the neon poster. Fibo was not flawless. It repeatedly added unwanted objects, put a stool in an aisle that was supposed to remain empty, duplicated a cylinder, and mangled crucial deployment text as “Paich.3 deployed.” Both models routinely ignored the requested 16:9 framing. Janus occasionally offered stronger drama, a clearer suggestion of a video call, or better preservation of an empty aisle, but those isolated advantages never became a task win. **Final call: Fibo Bbq Preview is the clear winner. Its errors are mostly local; Janus-Pro’s failures more often break the core spatial, anatomical, or compositional instruction.**
Bench layout precision
A photorealistic top-down product scene on a matte graphite workbench, lit by a soft overhead strip light, showing a folded cobalt drone controller exactly centered; a transparent amber microSD case is precisely 6 centimeters to the left of the controller; a silver hex screwdriver lies horizontally directly above it; a lime-green earbud charging case sits below it; a black smartwatch is to the right of the controller with its band open; a tiny brass SIM eject pin is placed between the smartwatch and the controller; and a white USB-C cable snakes behind the controller, partly obscured, all objects sharp, carefully spaced, premium catalog composition, 16:9
OpenAI: GPT-5.6 Sol Pro: Model B follows the requested spatial layout much more precisely, with a centered cobalt controller, amber case left, screwdriver above, green case below, smartwatch right, and cable routed behind. Model A has misplaced and duplicated objects, weaker top-down presentation, and less convincing object geometry; both models miss the requested 16:9 aspect ratio, while Model B’s SIM pin and microSD case are not fully accurate. (Second judge pass, order swapped — scores are the average of both: Model | Anthropic: Claude Fable 5.1: Model A is an angled, not top-down, view wi…
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.
OpenAI: GPT-5.6 Sol Pro: Model A includes multiple prohibited plants, a prominent lamp, and several framed artworks, while Model B cleanly respects all three negations and includes the requested chair, books, tea, window, and warm light. Model B is also polished and well balanced, though both outputs miss the requested 16:9 aspect ratio and Model B feels somewhat sparse rather than especially cozy. (Second judge pass, order swapped — scores are the average of both: Model B includes the requested armchair, books, tea, wi | Anthropic: Claude Fable 5.1: Model A violates every negation in the pro…
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.
OpenAI: GPT-5.6 Sol Pro: Model B correctly places the yellow duck on the matte green cube and preserves the main left/right and behind relationships, though it adds an extra blue cylinder and misses the requested 16:9 aspect ratio. Model A places the duck on the red sphere and does not clearly position the cylinder behind both objects, making its attribute binding substantially weaker. (Second judge pass, order swapped — scores are the average of both: Model B correctly binds the yellow duck to the green cube and places | Anthropic: Claude Fable 5.1: Model A places the duck on the red sphere…
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.
OpenAI: GPT-5.6 Sol Pro: Model B delivers much stronger, cleaner one-point perspective with highly consistent scale recession, symmetry, and warm lighting, though the foreground stool violates the empty-aisle requirement. Model A better preserves emptiness and appears somewhat more photographic, but its perspective, shelf geometry, detail, and sharpness are less convincing; both models miss the requested 16:9 aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B delivers much stronger | Anthropic: Claude Fable 5.1: Model B delivers a clean, centered one-poi…
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 B more clearly depicts a colorful friendship bracelet, with both hands anatomically convincing and a plausible grip, while Model A shows ambiguous plain string and obscures much of the finger positioning. Model B is also sharper and better composed, though both outputs miss the requested 16:9 aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B clearly depicts both anatomically plausible hands working with colorful friendship-bracelet threads in soft da | Anthropic: Claude Fable 5.1: Model B clearly shows a colorful braided f…
Neon tram poster
A clean vector-style vintage travel poster for the fictional tech district of Solena Quay, featuring a sleek autonomous tram gliding past a canal-side row of glowing repair kiosks and giant antenna sculptures at blue hour, with a cool teal, coral, and cream palette, crisp geometric shapes, minimal texture, elegant negative space, optimistic futuristic mood, and bold poster composition with the tram dominating the foreground, 16:9
OpenAI: GPT-5.6 Sol Pro: Model B more fully realizes the clean vector aesthetic, dominant foreground tram, glowing kiosk row, monumental antenna sculptures, and specified teal-coral-cream palette. Model A has appealing poster atmosphere but omits or minimizes several key district elements, renders the title incorrectly, and, like Model B, fails the requested 16:9 format. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the requested clean vector aesthetic, palette, canal-side g | Anthropic: Claude Fable 5.1: Model A includes text but misspells it ('S…
Courier leap with gadgets
A cinematic action illustration of a rooftop data courier in a saffron windbreaker leaping across a rain-slick alley at night while clutching a rugged transparent tablet that sprays droplets and throws cyan interface light onto their face, one sneaker barely leaving a bent metal vent, loose charging cables and delivery tags whipping behind, distant maglev lines streaking across the background, dramatic low-angle composition, intense motion blur on the edges, high-energy comic-realism, 16:9
OpenAI: GPT-5.6 Sol Pro: Model B more clearly depicts the transparent cyan-lit tablet, whipping delivery tags and cables, rain droplets, vent-assisted leap, and maglev-like background streaks with stronger technical polish. Model A offers a more dramatic low-angle comic composition, but the tablet is undersized and unclear, and both models miss the requested 16:9 format. (Second judge pass, order swapped — scores are the average of both: Model B captures substantially more prompt-specific details, especially the rugged | Anthropic: Claude Fable 5.1: Model B clearly renders the transparent tab…
Relief on video call
A close-up semi-realistic digital painting of a robotics engineer at a cluttered home desk, shoulders dropping in visible relief as a video call on a slim laptop confirms "Patch 7.3 deployed," their expression unmistakably relieved after tension: damp hair pushed back, eyes softening, mouth opening into a small exhausted smile, one hand pressed to the chest, warm monitor glow mixed with a dim tungsten desk lamp, shallow depth of field, intimate framing on face and upper torso, 16:9
OpenAI: GPT-5.6 Sol Pro: Model B more convincingly captures the intimate relief, damp hair, hand-to-chest gesture, robotics workspace, and mixed warm/cool lighting, with stronger composition and finish. Model A includes a clearer video-call element, but the laptop does not visibly confirm the deployment; Model B also loses adherence because its prominent text is incorrectly rendered as “Paich.3 deployed” rather than “Patch 7.3 deployed,” and neither image uses 16:9. (Second judge pass, order swapped — scores are the ave | Anthropic: Claude Fable 5.1: Model B delivers the relief beat convincin…
Matchup powered by OpenRouter.