Head to head: Wan vs Ideogram V4.5 Edit
Wan vs Ideogram V4.5 Edit
By Ryan Merket · Published
Eight image-generation tasks put composition, exact counts, spatial relationships, and prompt constraints under the microscope. The split was not close: one model won every task, from a quiet reading nook to a storm-lit fantasy sprint.
Wan takes this matchup by a landslide: **66.8 to 47.7, with a tentative lean and an 8–0 task record.** The decisive edge was not just polish; Wan repeatedly obeyed the brief where Ideogram V4.5 Edit drifted from it. The misses were concrete. In the negation test, Ideogram added the very plants, lamp, and framed art the prompt excluded. In the void joust, its lance pointed behind the rider, and the storm-lit chasm gave way to a less faithful setting. Wan delivered the requested action and atmosphere in both. Wan also proved more dependable on structured scenes. It rendered exactly seven cups, placed the duck and colored objects in the specified arrangement, and maintained the requested perspective and wide framing. For the alchemist’s desk, it got the noir palette and object placement right; for the nebula reliquary, it showed 11 distinct orbs rather than Ideogram’s crowded, linked-looking cluster. Ideogram can produce polished images, but this set exposes a recurring weakness: it often treats explicit constraints—especially aspect ratio, counts, and spatial instructions—as optional. **The verdict: Wan wins decisively, not on style alone, but because it followed the prompts far more reliably across all eight tests.**
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-6 Luna: Model A closely matches the requested nook and avoids indoor plants, lamps, and artwork, though greenery is visible outside the window. Model B includes multiple plants, a lamp, and framed artwork, and it is square rather than 16:9. (Second judge pass, order swapped — scores are the average of both: Model B is warm and polished, but prominently includes plants, a lamp, and framed artwork, and it is not 16:9. Model A captures the cozy reading nook with books, tea, and afternoon light while avoidi | Anthropic: Claude Fable 5.1: Model A fully respects the negation constraints…
Void joust sprint
A fantasy action scene in painterly high-detail style: a moon-pale knight of the Ash Banner riding a six-legged obsidian elk mid-leap across a broken bridge over a glowing chasm, cloak snapping violently backward, sparks spraying from shattered stone under the hind legs, lance thrust forward toward an unseen foe, the animal’s muscles extended and hooves blurred with speed, debris and embers trailing through the air to create a strong sense of motion; storm-lit sky, dramatic side lighting, dynamic diagonal composition, 16:9.
OpenAI: GPT-6 Luna: Model A more fully captures the storm-lit sprint over a glowing chasm, with a forward-thrust lance, sparks, and a dynamic 16:9 composition. Model B has a striking antlered mount and polished rendering, but its lance points behind the rider and it misses much of the stormy, void-lit setting. (Second judge pass, order swapped — scores are the average of both: Model B captures the antlered mount, broken bridge, and flying sparks, but its square framing and backward-pointing lance weaken the request | Anthropic: Claude Fable 5.1: Model A delivers the 16:9 format, storm-lit lig…
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 exactly seven clearly separate, identical-looking espresso cups in a straight row on white marble and matches the requested 16:9 framing. Model B also has seven cups in a row, but its square framing, stronger shadows, and less uniform espresso appearance make it a weaker match. (Second judge pass, order swapped — scores are the average of both: Model A presents exactly seven clearly separated, highly uniform cups in a clean straight row, making the count and arrangement especially | Anthropic: Claude Fable 5.1: Model A shows exactly seven identical white espr…
Rain-soaked maglev canyon
A futuristic cityscape at dusk in a realistic concept-art style: viewed from the middle of a rain-slick elevated maglev line running straight into a single clear vanishing point between colossal tower-blocks, with correct perspective and scale throughout—near foreground maintenance drones large and detailed, mid-distance commuter pods smaller, and far-off skybridges tiny against the haze; nothing warped, no impossible angles, reflections consistent with the architecture; cold cyan signage, wet asphalt sheen, long-lens cinematic composition, 16:9.
OpenAI: GPT-6 Luna: Model A more fully captures the rain-slick elevated maglev canyon, with cyan reflections, detailed foreground maintenance vehicles, smaller commuter pods, and a clear central vanishing point. Model B has a strong symmetrical city corridor, but its square framing, less evident rain, and less distinct foreground-to-distance scale make it less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model A more fully captures the rain-soaked, wide cinematic magle | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 long-lens c…
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-6 Luna: Model A closely matches the requested photorealistic 16:9 tabletop scene, with the duck atop the green cube, the red sphere to its right, and the blue cylinder behind them. Model B preserves the object colors and arrangement but looks more like a stylized render and uses a square composition rather than 16:9. (Second judge pass, order swapped — scores are the average of both: Model A more closely matches the photorealistic tabletop brief and requested 16:9 framing, with all four objects clearly | Anthropic: Claude Fable 5.1: Model A delivers a photorealistic 16:9 tabletop…
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-6 Luna: Model A more closely matches the requested 16:9 photorealistic aisle, with warm overhead lighting, clear one-point perspective, and consistent diminishing scale. Model B has a strong centered vanishing point, but its square framing and softer, less warm rendering make it a weaker match. (Second judge pass, order swapped — scores are the average of both: Model A more closely matches the requested 16:9 framing and presents a clean, convincing one-point perspective with consistently receding shelve | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 frame, warm…
Alchemist desk layout
A film-noir black-and-white image of a clandestine arcane laboratory desk, lit by a single hard overhead lamp: a glass cube containing a tiny storm cloud sits exactly between a tarnished silver dagger on the left and a closed velvet-bound grimoire on the right; behind the cube stands a brass astrolabe; on top of the grimoire rests a white porcelain raven mask; in front of the cube lies a key with three teeth angled toward the lower-right corner; a sealed vial of dark liquid is placed behind the dagger but left of the astrolabe; precise spatial arrangement, clean tabletop composition, crisp shadows.
OpenAI: GPT-6 Luna: Model A follows the requested monochrome noir lighting and precise object placement more closely, including the cube between the dagger and grimoire, with the vial behind the dagger and the key angled toward the lower-right. Model B is polished and gives the mask a clearer raven shape, but its colored palette and key orientation depart from the prompt. (Second judge pass, order swapped — scores are the average of both: Model A more closely matches the monochrome noir lighting and the requested o | Anthropic: Claude Fable 5.1: Model A delivers the required film-noir black-a…
Nebula reliquary count
A baroque sci-fi still life inside the reliquary vault of the starship Ossuary of Nhal, painted as an ultra-detailed cinematic digital illustration: exactly 11 distinct levitating amber memory-orbs arranged in a loose spiral above a cracked obsidian altar, each orb fully visible and separately countable, plus etched brass instruments and drifting incense smoke in the background that do not resemble extra orbs; dramatic rim lighting from violet reactor slits, deep shadows, sharp focus, centered composition, 16:9.
OpenAI: GPT-6 Luna: Model A clearly shows exactly 11 separately countable amber orbs above a cracked altar, with violet lighting, smoke, brass instruments, and the requested wide cinematic framing. Model B is polished but presents the orbs as a connected bead-like loop rather than distinct levitating objects, and the count is not 11. (Second judge pass, order swapped — scores are the average of both: Model B is polished but presents the orbs as a crowded, wire-linked ring and appears to exceed the requested count. | Anthropic: Claude Fable 5.1: Model A delivers exactly 11 clearly separated am…
Matchup powered by OpenRouter.