Head to head: Ideogram V4.5 Edit vs Kling Image
Ideogram V4.5 Edit vs Kling Image
By Ryan Merket · Published
A practical image-model matchup across eight demanding tests, from perspective and object counts to style fidelity, anatomy, and prompt exclusions. The decisive question is which model stays reliable when a brief asks for more than an attractive image.
Ideogram V4.5 Edit takes the matchup, 61.4 to 59.1, with an limited confidence verdict and a 5–2–1 task record. That’s a clear result across the test set, not a blowout: Kling Image wins two specific tasks, and the aggregate gap remains modest. Ideogram’s edge comes from handling briefs with several constraints at once. It wins on Bauhaus style fidelity, showing a more disciplined poster language and cleaner wayfinding; on anatomy, where the subject visibly handles both a blueprint and caliper; and on the two negation tests. In the archival hall, Kling’s visible ladder and chairs violate the exclusions, while Ideogram better preserves the requested atmosphere. For the reading nook, Ideogram also makes the scene feel more cohesive. Kling is stronger when composition or exactness is the central demand. Its bookstore aisle has cleaner centered one-point perspective, and its seven espresso cups read as a more uniform, clearly separated row. The duck-on-cube attribute-binding task is a tie: both get the arrangement and colors right, but each has a compositional drawback. The pattern is telling: Kling can nail a narrowly defined visual target, but Ideogram more often keeps the whole prompt intact. **Verdict: Ideogram V4.5 Edit wins—clearly on the results, though not by a landslide.**
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 B presents the clearest centered one-point perspective, with the shelves, floor, and overhead lights converging consistently into a long aisle. Model A is atmospheric and convincing too, but the ladder draws attention from the empty aisle and makes the composition less clean. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the warm, empty one-point-perspective aisle, with clear, consistent shelf and book scale receding to a centered vanishin | Anthropic: Claude Fable 5.1: Model A delivers a convincingly photorealistic…
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 more closely follows the requested soft studio look and keeps the duck small while clearly placing it atop the green cube; Model B has the correct object arrangement but makes the duck disproportionately large and includes distracting studio equipment in the background. Both images render the colors and spatial relationships well, though Model A’s cleaner composition is more polished. (Second judge pass, order swapped — scores are the average of both: Model A presents the requested color | Anthropic: Claude Fable 5.1: Both bind colors correctly and place the duck o…
Named art-style fidelity
A faithful Bauhaus poster-style depiction of the interior lobby of the Meridian Pneumatics House, 16:9: double-height space with tubular steel reception desk, circular pendant lights, a red stair tower, black-and-cream wall planes, and a single standing coat rack; strict asymmetrical composition, flat primary colors plus black and beige, precise geometric forms, sans-serif wayfinding text blocks, minimal shading, and the disciplined graphic language of classic Bauhaus design.
OpenAI: GPT-6 Luna: Model A more faithfully captures the disciplined, flat Bauhaus poster language, with a strong asymmetrical layout, primary-color accents, circular pendants, and correctly rendered wayfinding. Model B includes a clearer stair tower and tubular reception desk, but its heavier dimensional rendering and misspelled signage weaken the match. (Second judge pass, order swapped — scores are the average of both: Model A has the more disciplined Bauhaus poster language, strong asymmetry, clean geometric pl | Anthropic: Claude Fable 5.1: Model A delivers a disciplined Bauhaus poster l…
Human anatomy & hands
A clear full-body documentary-style portrait, 16:9, inside the sunlit atrium of the Kestrel & Morrow Tile Conservatory: a restoration architect stands on a terrazzo walkway beside scale models and sample panels, facing camera at a slight angle, natural body proportions, both hands fully visible and unobstructed as she carefully holds an unfolded blueprint with one hand and a brass caliper with the other, all ten fingers anatomically correct, sleeves rolled to the forearms, crisp late-morning window light, balanced realistic composition.
OpenAI: GPT-6 Luna: Model A more closely follows the key action, visibly holding both the blueprint and caliper in the conservatory, though the subject is kneeling rather than standing. Model B has a polished, balanced composition, but the caliper is not in hand and the framing is not full-body. (Second judge pass, order swapped — scores are the average of both: Model A more clearly shows both hands handling the blueprint and caliper, with the tile models and sunlit atrium present; Model B has a stronger standing, | Anthropic: Claude Fable 5.1: Model A delivers the blueprint-in-one-hand, cali…
Stylized illustration
Clean vector illustration, 16:9: a narrow interior courtyard of the fictional Hotel Virelai after a summer storm, with apricot stucco walls, dark green steel balconies, striped awnings dripping rainwater, potted lemon trees, and a small tiled fountain in the center; mood is calm and slightly wistful, with soft twilight reflections in puddles, simplified geometric shapes, sharp edges, restrained grain-free color fields, and elegant poster-like composition.
OpenAI: GPT-6 Luna: Model A better captures the calm, wistful post-storm twilight with reflective puddles, apricot walls, balconies, lemon trees, and a central tiled fountain. Model B has clearer striped awnings and a bold poster-like style, but its intense contrast and simplified setting feel less like the requested soft twilight courtyard. (Second judge pass, order swapped — scores are the average of both: Model B delivers a crisp, poster-like vector style and includes the key courtyard elements, but its oversize | Anthropic: Claude Fable 5.1: Model A delivers a coherent narrow interior cou…
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 B shows exactly seven clearly separated, nearly identical espresso cups in a straight row on white marble. Model A also has seven cups, but the saucers and inconsistent, spoon-like handles make the cups look less identical and the arrangement less clean. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly seven clearly separated, matching espresso cups in a straight row on white marble, with a clean overhead composition. Model A also has seven cups, bu | Anthropic: Claude Fable 5.1: Model A shows exactly seven fully visible, even…
Negation & exclusion
Documentary photojournalism style, 16:9: the reading room of the defunct Brine Ledger Exchange in Port Serevan, a long walnut hall with brass desk lamps, teal accession tags, rain-streaked clerestory windows, and a mosaic floor showing a compass rose; three archivists in slate cardigans sort stacks of oversized maritime ledgers under cool overcast daylight, with dust visible in the air; the room must be completely WITHOUT any chairs, stools, ladders, or other seating anywhere in frame.
OpenAI: GPT-6 Luna: Model A more closely captures the active archival work and atmospheric hall, though it appears to include an extra archivist and possibly seating in the background. Model B has three archivists but visibly includes a ladder and chairs, directly violating the exclusion requirement. (Second judge pass, order swapped — scores are the average of both: Model B captures the long walnut hall, three archivists, teal tags, and compass mosaic, but a ladder is clearly visible, violating the prompt’s explic | Anthropic: Claude Fable 5.1: Model A keeps the frame free of chairs, stools,…
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 more fully evokes a cozy reading nook with an open book, stacked books, steaming tea, and warm window light, while respecting all the exclusions. Model B is attractive and also avoids plants, lamps, and artwork, but the mug on the floor makes the scene feel less like a finished reading nook. (Second judge pass, order swapped — scores are the average of both: Both images include the requested nook elements, warm window light, and avoid plants, lamps, and artwork. Model A feels more like a | Anthropic: Claude Fable 5.1: Model A delivers a convincing, cohesive nook wi…
Matchup powered by OpenRouter.