Head to head: Ideogram V4.5 Edit vs Seedream 5.0 Pro Image Editing
Ideogram V4.5 Edit vs Seedream 5.0 Pro Image Editing
By Ryan Merket · Published
Eight image-editing challenges put composition, aspect-ratio control, object placement, and prompt constraints under pressure. The results show which model more reliably turns detailed instructions into a finished frame—and where visual polish alone isn’t enough.
Seedream 5.0 Pro Image Editing wins decisively: **67.2 to 45.0, taking all eight tasks with 100% statistical confidence.** The gap reflects a consistent advantage in following the brief, not one standout image carrying the matchup. The clearest pattern is format discipline. Seedream repeatedly delivered the requested 16:9 frame, from the campsite and warehouse aisle to the rooftop workshop and macro study. Ideogram often produced polished images, but square framing undercut its fit to the brief—even when its composition was otherwise strong. Seedream also handled specific content and exclusions more reliably. It showed the requested carnivorous plants and kiosk details more accurately, and its reading nook avoided the forbidden plants, lamps, and artwork that Ideogram included. Ideogram did have a notable edge in the clean isometric bedroom layout, while Seedream’s plant grid cropped pots and its palette was not strictly limited to four colors. Those misses matter, but they didn’t outweigh Seedream’s broader task adherence. **Final call: Seedream 5.0 Pro Image Editing is the clear winner for prompt-faithful editing. Ideogram can deliver a polished image, but across this set it too often missed a defining constraint—especially the requested frame shape—while Seedream more consistently got the scene and its requirements right.**
Restricted palette
A flat-vector illustration of a mountain campsite at dusk using ONLY four colors — cream, burnt orange, teal, and charcoal — with no gradients and no stray off-palette colors, balanced composition, 16:9.
OpenAI: GPT-6 Luna: Model B matches the requested 16:9 format and presents a clear, polished mountain campsite, while Model A is a strong illustration but square rather than widescreen. Both use flat vector styling, though neither strictly limits itself to only four distinct color values. (Second judge pass, order swapped — scores are the average of both: Model B better meets the requested 16:9 format and presents a clear, balanced campsite scene in a mostly restricted four-color palette. Model A is polished, but i | Anthropic: Claude Fable 5.1: Model A violates the brief with a 1:1 aspect ra…
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 better matches the requested 16:9 format and has a convincing, photorealistic one-point-perspective aisle with warm overhead lighting. Model A has a particularly clean, centered vanishing point, but its square framing misses the specified aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested 16:9 framing while preserving a clear, centered one-point perspective, warm lighting, and convincing scale progression. Model A also h | Anthropic: Claude Fable 5.1: Model B delivers the requested 16:9 frame with…
Nine Carnivorous Pots
Clean top-down flat lay product photograph on a pale basalt tabletop, 16:9, showing EXACTLY nine distinct potted carnivorous plants arranged in a precise but natural grid with clear separation so every item is individually countable: two Venus flytraps, three pitcher plants of different species, one sundew, one butterwort, one cobra lily, and one bladderwort in a glass bowl; each pot has unique shape and mulch texture, tiny handwritten zinc labels, soft diffused greenhouse light, crisp shadows, and no extra plants, cuttings, reflections, or duplicated leaves that could confuse the count.
OpenAI: GPT-6 Luna: Model A has a cleaner overhead grid with all nine pots fully visible, but several plants do not clearly match their labeled species and the image is not 16:9. Model B depicts the requested plant assortment more accurately and uses the requested wide format, though the bottom two pots are cropped and the view is less top-down. (Second judge pass, order swapped — scores are the average of both: Model B presents all nine requested plants with mostly clear species labels and distinct pots, while Mod | Anthropic: Claude Fable 5.1: Model B delivers the 16:9 format, nine clearly…
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 more closely follows the requested clean isometric, flat-vector layout, with the bed, centered rug, desk beneath the back-wall window, and lamp in the front-right corner clearly placed. Model B is attractive and includes the requested furnishings, but its perspective is less consistently isometric and its rug is pushed toward the foreground. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested placements and delivers a polished, spacious isome | Anthropic: Claude Fable 5.1: Model A places the bed, desk-under-window, and…
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 multiple forbidden plants, a lamp, and framed artwork, and it misses the requested 16:9 format. Model B includes the reading nook essentials in warm afternoon light while avoiding all the prohibited items and matching the wide format. (Second judge pass, order swapped — scores are the average of both: Model B captures the requested cozy nook with an armchair, books, tea, and warm afternoon window light, while avoiding plants, lamps, and artwork. Model A is visually appealing but | Anthropic: Claude Fable 5.1: Model A violates every negation in the prompt —…
Terrarium Kiosk Reflections
A twilight editorial photograph, 16:9, of a narrow plant-repair kiosk called "Moss & Mercury" inside a rainy transit arcade: a cracked brass-framed terrarium on the counter, panes beaded with condensation and tiny rivulets, reflects the neon timetable opposite while also showing the refracted stems of miniature pilea, purple oxalis, and a tiny ceramic snail inside; behind it, a chrome watering can, a fogged mirror tile, and wet black floor tiles must all mirror the surrounding scene consistently from their angles, with believable transparency through the glass and realistic distortion in the curved surfaces, lit by cyan-pink station signage and one warm tungsten task lamp.
OpenAI: GPT-6 Luna: Model B better matches the requested 16:9 editorial scene, with the kiosk, timetable, rain-beaded terrarium, plants, watering can, and wet-floor reflections all present. Model A has a striking centered composition and clear terrarium details, but misses the kiosk identity and requested wide format. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested widescreen editorial scene more closely, with the cracked brass terrarium, visible condensation, minia | Anthropic: Claude Fable 5.1: Model B delivers the 16:9 aspect, the legible '…
Macro Dew Lens Study
Hyper-detailed botanical macro photograph of a variegated thyme cutting on dark slate, shot inches away with a true macro lens at f/2.8 so only one serrated leaf edge and a single dew droplet are razor sharp while the background seed packets and terracotta rim dissolve into creamy circular bokeh; the droplet acts like a tiny lens showing an inverted miniature greenhouse frame inside it, with early-morning side light, soft shadow falloff, and a cinematic 16:9 crop.
OpenAI: GPT-6 Luna: Model B better matches the requested cinematic wide crop, with blurred seed packets and a dew droplet resting on a thyme leaf; Model A has a more striking, crisp macro look, but its oversized droplet sits on the slate and too much of the cutting is in focus. Model B’s focal detail is somewhat soft, while Model A’s greenhouse-like reflection is more legible. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested cinematic 16:9 macro photograph, wi | Anthropic: Claude Fable 5.1: Model B delivers the 16:9 cinematic crop, genui…
Rooftop Grafting Workshop
Painterly contemporary illustration, 16:9, of a windswept rooftop gardening workshop at blue hour: five distinct people collaborate around a long reclaimed-wood table in one balanced composition—an elderly woman tying a graft on a dwarf pear sapling, a teenager holding a tablet with growth notes, a courier in a saffron rain cape unloading trays of sea kale seedlings, a tattooed botanist adjusting a copper misting nozzle, and a child kneeling beside a crate of striped pumpkins feeding a rescue pigeon—surrounded by orderly clusters of tools, seedling flats, lanterns, and skyline planters, with coherent eyelines, overlapping gestures, and believable shared space under strings of warm festival bulbs.
OpenAI: GPT-6 Luna: Model B better matches the requested 16:9 painterly scene and shows all five collaborators in a believable, connected workshop, including the grafting, copper nozzle, pumpkins, and pigeon. Model A includes most of the specified elements, but its square framing and more isolated arrangement weaken the fit to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B more fully captures the requested painterly blue-hour rooftop workshop, with clearly differentiated col | Anthropic: Claude Fable 5.1: Model B delivers a true 16:9 painterly blue-hou…
Matchup powered by OpenRouter.