Head to head: Krea 3 Realtime vs Ideogram V4.5 Edit
Krea 3 Realtime vs Ideogram V4.5 Edit
By Ryan Merket · Published
Eight image tasks put composition, visual polish, prompt adherence, and small details under the microscope. The sharpest contrasts came from following exact framing and content constraints versus producing a cleaner, more controlled-looking image.
Krea 3 Realtime wins this matchup 62.9 to 50.2, with a tentative lean: a decisive result across eight tasks, not a narrow call. It took seven task wins to Ideogram V4.5 Edit’s one, and its advantage came largely from doing what the prompts specifically asked for—not simply making attractive images. Krea was more reliable on the demanding constraint-heavy prompts. It kept the requested 16:9 framing and details in the reflections, Art Deco poster, tabletop, and palette tasks, and it respected the exclusions in the negation prompt. Ideogram’s polished images repeatedly missed those specifics: several were square, and its cozy-room scene included plants, a lamp, and artwork despite being told to omit them. Ideogram’s win came in spatial layout, where its flat-vector treatment placed the desk and front-right lamp more convincingly. It also had the more legible hands in the anatomy task, with a clearer bracelet grip; Krea better conveyed the act of tying and matched the wide framing, but its fingers were harder to read and the image showed texture artifacts. That’s a real visual advantage for Ideogram, but not enough to outweigh Krea’s broader prompt compliance. **Final call: Krea 3 Realtime is the stronger pick in this head-to-head—more dependable when the brief specifies framing, objects, and what must stay out of the image. Ideogram V4.5 Edit can look cleaner, but polish alone didn’t carry the score.**
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-6 Luna: Model B shows clearer, more anatomically convincing hands with a plausible grip on the bracelet, and has stronger overall image quality. Model A better conveys active tying and matches the requested wide framing, but the clenched grips obscure the fingers and the image has visible texture artifacts. (Second judge pass, order swapped — scores are the average of both: Model B shows both hands clearly pinching a recognizable friendship bracelet with more convincing anatomy and a natural grip, thoug | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 candid fram…
Spatial layout
A clean isometric illustration of a bedroom: a bed against the LEFT wall, a round rug centered on the floor, a desk under the WINDOW on the back wall, and a floor lamp in the FRONT-RIGHT corner. Flat-vector style, consistent perspective.
OpenAI: GPT-6 Luna: Model A follows the requested layout closely, but Model B has a more polished, legible flat-vector style and places the floor lamp more convincingly in the front-right area. Model B’s rug sits somewhat forward of center, while Model A’s rug is closer to the requested centered placement. (Second judge pass, order swapped — scores are the average of both: Model B more clearly places the desk beneath the window and the floor lamp in the front-right corner, while also delivering a polished, consiste | Anthropic: Claude Fable 5.1: Model A places the bed on the left wall and cen…
Perspective & scale
A stop-motion claymation style product photo of the interior of the fictional Rill & Fable flagship luggage shop, viewed from knee height down a long aisle in strict one-point perspective, 16:9: a tiny mint-green keychain suitcase sits on the front edge of a display table at left, behind it three carry-on suitcases, then farther back four full-size checked cases, and at the very end of the aisle a towering wall display of stacked trunks under a skylight; every suitcase must have believable relative size based on distance, parallel shelf lines converging cleanly to a single vanishing point, floor tiles and ceiling slats aligned with that same perspective, and no warped handles, stretched wheels, or wrongly scaled price tags; warm morning retail lighting, crisp shadows, tactile handmade clay textures.
OpenAI: GPT-6 Luna: Model A more closely follows the requested scale progression, especially with the tiny mint-green keychain suitcase at the front left and smaller cases set farther down the aisle. Model B has cleaner, stronger one-point symmetry, but its large foreground suitcases omit the keychain and disrupt the requested distance-based sequence. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested wide aisle scene and includes the tiny mint suitcase on a fro | Anthropic: Claude Fable 5.1: Model A delivers nearly every requirement: 16:9…
Reflections, glass & transparency
A cinematic close-up of a transparent electric kettle and two double-walled glass mugs from the fictional homewares brand North Spindle on a polished chrome countertop, 16:9, in a quiet early-evening kitchen; the kettle is half full, showing the refracted distortion of a striped tea towel behind it, one mug contains amber tea with a visible meniscus, the other is empty and catches reflections from a small tangerine and a cobalt timer nearby, and a brushed steel toaster appears correctly mirrored in the chrome surface; all glass, water, and metal reflections must match object positions and angles accurately, with believable transparency, refraction, highlights, and layered reflections under warm under-cabinet lighting.
OpenAI: GPT-6 Luna: Model A includes the refracted striped towel, tea and empty glass mugs, tangerine, cobalt timer, chrome reflections, and the requested wide cinematic framing. Model B is polished, but omits the striped towel and is square rather than 16:9, with less convincing fulfillment of the specific reflection details. (Second judge pass, order swapped — scores are the average of both: Model A more directly fulfills the requested widescreen scene and clearly shows the striped towel refracted through the ket | Anthropic: Claude Fable 5.1: Model A delivers the 16:9 framing, a convincing…
Named art-style fidelity
An Art Deco advertising poster for a fictional consumer product, the Vestra No. 7 countertop espresso machine, shown as the central hero object on a stepped pedestal in a glamorous hotel-lobby vignette, 16:9; the image must faithfully embody Art Deco with streamlined geometry, sunburst motifs, chevrons, symmetrical composition, luxurious lacquered surfaces, stylized smoke plumes, and a refined palette of black, ivory, brass, and emerald; dramatic spotlighting, crisp graphic edges, elongated ornamental forms, and period-authentic poster design language rather than generic vintage art.
OpenAI: GPT-6 Luna: Model A more fully reads as a period-authentic 16:9 advertising poster, with clear Vestra No. 7 typography, a symmetrical hotel-lobby setting, and a refined black, emerald, ivory, and brass palette. Model B has bold Art Deco sunbursts and a strong pedestal display, but its square format and less prominent product advertising make it a weaker match overall. (Second judge pass, order swapped — scores are the average of both: Model A more faithfully reads as a period-authentic Art Deco advertising | Anthropic: Claude Fable 5.1: Model A delivers a true 16:9 Art Deco poster wit…
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 matches the requested 16:9 tabletop scene, with the specified colors and object placements clearly readable. Model B is polished and follows the object relationships well, but its square framing and muted teal-green cube are less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested 16:9 photorealistic tabletop scene, with the duck on the green cube and the red sphere in front of the blue cylinder. Mo | Anthropic: Claude Fable 5.1: Model A satisfies every constraint—16:9 framing…
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 includes the requested armchair, books, tea, window, and warm afternoon light while avoiding plants, lamps, and artwork. Model B is visually inviting but prominently includes all three forbidden categories and is not in the requested 16:9 format. (Second judge pass, order swapped — scores are the average of both: Model B is visually inviting but prominently includes plants, a lamp, and framed artwork, violating all three explicit exclusions and the requested aspect ratio. Model A omits t | Anthropic: Claude Fable 5.1: Model A correctly follows the negation constrai…
Color-palette accuracy
A meticulously staged hero shot of six everyday skincare bottles from the invented brand Luma Thread arranged on a bathroom shelf in a soft editorial still life, 16:9, rendered in smooth claymation style under diffused north-window light; the entire image must be restricted to the named Dawnberry palette only: alabaster, dusty mauve, muted persimmon, sage gray, deep plum, and brushed cocoa, with absolutely no other hues anywhere in the bottles, labels, shelf, towels, shadows, reflections, or background steam; elegant balanced composition, subtle material contrast, premium consumer-product styling.
OpenAI: GPT-6 Luna: Model A closely follows the restricted Dawnberry palette, uses the requested 16:9 claymation still-life presentation, and keeps even the shelf and background within the muted tones. Model B has a polished arrangement, but its black pump hardware, warmer lighting, and square format depart more noticeably from the prompt. (Second judge pass, order swapped — scores are the average of both: Model A follows the requested 16:9 framing and uses the Dawnberry palette more consistently across the bottles | Anthropic: Claude Fable 5.1: Model A delivers a true claymation render in 16…
Matchup powered by OpenRouter.