Head to head: Hidream O1 Image vs Qwen Image 3 Image Editing
Hidream O1 Image vs Qwen Image 3 Image Editing
By Ryan Merket · Published
This matchup tests whether visual polish can survive exacting demands around counting, spatial relations, exclusions, typography, style, and complex scene construction. The decisive factor is not beauty in isolation, but disciplined prompt execution.
Qwen Image 3 Image Editing won this matchup outright: **69.8 to 43.4**, with a clear verdict at **limited confidence**. It swept all **24 task decisions**, leaving Hidream O1 Image with no wins and no ties—a level of consistency that rules out a narrow or category-dependent advantage. The separation was most obvious when prompts required literal compliance. Qwen placed the blue cylinder behind both specified objects, excluded every forbidden plant, lamp, and framed artwork from the reading nook, and produced exactly thirteen countable watering cans. Hidream repeatedly traded accuracy for attractiveness: its scenes could look polished while breaking spatial instructions, missing required objects, or directly violating negations. Qwen also handled dense compositions and stylistic constraints better. Its rooftop seed swap preserved the requested cast, actions, props, and legible packet label; its orchid house included the transparent propagation table, visible roots, rain, mirrors, and chrome reflections; and its fishing scene read as ukiyo-e rather than a modern illustration borrowing the surface vocabulary. Even in the library-perspective task, where Hidream showed strong geometry, Qwen paired convincing recession with markedly better photorealism and detail. **Final call: Qwen Image 3 Image Editing wins decisively. Hidream O1 Image can make appealing pictures, but Qwen is the substantially more reliable model when prompts demand exact counts, bindings, exclusions, text, style fidelity, and multi-element scene logic.**
Rooftop Seed Swap
A cinematic editorial photograph, 16:9, of a twilight rooftop community seed-swap garden in early spring: seven distinct people interacting in one coherent scene — a gray-haired woman kneeling to repot lemon thyme, a teenager in a mustard raincoat handing her a packet labeled 'Blue Fen Parsnip', a man on a ladder hanging copper lanterns over a trellis of sweet peas, two children crouched together releasing ladybugs onto kale leaves, a florist in a plum apron arranging cut dahlias on a crate table, and a sleepy border collie lying beside stacked terracotta pots — with a striped greenhouse wall, watering cans, seed trays, and string lights balanced across the composition; soft golden-hour light with cool city buildings beyond, natural poses, clear spatial relationships, and believable interaction among all subjects.
OpenAI: GPT-5.6 Sol Pro: Model B captures nearly every specified action and prop, including the legible seed-packet label, ladybugs, thyme, lanterns, plum-apron florist, stacked pots, and cohesive rooftop setting, with excellent spatial storytelling. Model A is attractive but misses or weakens several key details, has less convincing subject roles and interaction, and does not render the requested packet text clearly. (Second judge pass, order swapped — scores are the average of both: Model B closely realizes the crowde | Anthropic: Claude Fable 5.1: Model B renders nearly every requested ele…
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 blue cylinder behind both the green cube and red sphere while preserving all object-color bindings and the duck’s position. Model A is more richly photorealistic, but its cylinder sits in front of the cube rather than behind both objects. (Second judge pass, order swapped — scores are the average of both: Model B precisely preserves the requested colors and spatial relationships, including placing the blue cylinder behind both main shapes, while maintaining a polishe | Anthropic: Claude Fable 5.1: Model A gets colors and left/right placeme…
Named art style
A ukiyo-e woodblock print of a fishing boat riding a large cresting wave at dawn, faithful to the flat color planes, bold outlines, and stylized foam of the tradition, muted indigo and cream palette.
OpenAI: GPT-5.6 Sol Pro: Model B more faithfully captures a traditional ukiyo-e fishing scene through its convincing crewed boat, dawn sun, dramatic cresting wave, restrained indigo-and-cream palette, and finely stylized foam. Model A has a strong wave silhouette but weaker dawn cues, less convincing fishing activity, coarser details, and visibly garbled inscription-like text. (Second judge pass, order swapped — scores are the average of both: Model B more faithfully captures the requested dawn setting, fishing boat, mu | Anthropic: Claude Fable 5.1: Model B delivers a convincing ukiyo-e prin…
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 B closely follows every requested element and prohibition while presenting a polished, warmly lit, well-balanced reading nook. Model A is attractive and technically strong but directly violates all three negations by including multiple plants, a lamp, and framed artwork. (Second judge pass, order swapped — scores are the average of both: Model B fully honors the exclusions while presenting the requested armchair, books, tea, window, warm afternoon light, and cozy 16:9 composition. Model A | Anthropic: Claude Fable 5.1: Model A violates every negation in the prom…
Thirteen Watering Cans
A detailed watercolor illustration of a quirky gardener's workbench inside a potting shed, lit by crisp morning window light, showing EXACTLY thirteen distinct watering cans, all fully visible and individually countable: 1 tall zinc can with a long spout, 1 tiny red toy can, 2 chipped cream enamel cans, 3 matte sage-green cans of different shapes, 1 translucent amber plastic can, 1 hammered copper can, 2 cobalt-blue cans, 1 black square-bodied can, and 1 striped yellow-and-white can; arranged on shelves and the bench with seed packets, twine, moss, and labeled herb cuttings around them, but no extra watering cans, no partial duplicates, and no obscured cans.
OpenAI: GPT-5.6 Sol Pro: Model B clearly presents exactly thirteen fully visible, individually countable watering cans with the requested colors, materials, shapes, and surrounding gardening details, while Model A shows far fewer cans and misses several required variants. Model B also has a stronger watercolor character and well-organized shelving, though some small label text is imperfect. (Second judge pass, order swapped — scores are the average of both: Model B clearly presents exactly thirteen individually countabl | Anthropic: Claude Fable 5.1: Model B renders exactly thirteen countable…
Orchid House Refractions
A hyper-realistic image of a Victorian orchid house interior after rain, 16:9, centered on a long glass propagation table beneath a tall antique mirror and beside a chrome misting canister; the scene must show accurate reflections and transparency: raindrops on the greenhouse panes bending the view of hanging ferns outside, the mirror reflecting the back of a striped ceramic orchid pot and part of the arched roof from the correct angle, the glass tabletop revealing roots, labels, and pruning shears beneath it with realistic distortion, and the polished chrome canister reflecting nearby pink moth orchids and window frames; cool overcast daylight with small warm lamps, intricate highlights, believable optics, and no impossible duplicate objects in reflections.
OpenAI: GPT-5.6 Sol Pro: Model B much more fully realizes the rainy Victorian orchid house, long transparent propagation table, arched architecture, warm lamps, layered plants, mirror, and reflective chrome, with stronger atmosphere and optical detail. Model A is clean but sparse, lacks convincing rain and visible under-table roots and labels, and its mirror reflection does not clearly show the requested back of the striped pot. (Second judge pass, order swapped — scores are the average of both: Model B closely captures | Anthropic: Claude Fable 5.1: Model B delivers a convincing Victorian or…
Color-Bound Potting Shelf
A clean studio-style still life, 16:9, of a greenhouse potting shelf with eight separate objects, each required to keep its own exact color and material with no mixing: a glossy cerulean ceramic watering jug, a matte terracotta pot, a transparent emerald glass spray bottle, a brushed brass hand trowel, a lavender-painted wooden seed box, a white marble plant label marker, a charcoal rubber gardening glove, and a crimson velvet cushion holding cactus seeds; arranged in a balanced row with a few neutral green seedlings in the background only, bright diffused daylight, sharp focus, minimal backdrop, and strict color-to-object binding.
OpenAI: GPT-5.6 Sol Pro: Model B presents all eight objects with clearer material and color binding, a more balanced row, convincing textures, and sharper studio polish. Model A includes the required set but has awkward overlap and cropping, less convincing marble and wood cues, an unusually shaped marker, and illegible label text. (Second judge pass, order swapped — scores are the average of both: Model B clearly presents all eight requested objects with strong color-to-object and material binding, a balanced row, shar | Anthropic: Claude Fable 5.1: Model B renders all eight objects with cor…
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 a more convincing long, empty one-point-perspective aisle with consistent recession, rich photorealistic detail, and attractive warm lighting. Model A is symmetrical and on-prompt but appears flatter and more synthetic, with less natural shelf spacing and book detail. (Second judge pass, order swapped — scores are the average of both: Model B delivers a more convincing photorealistic library aisle with a clear central vanishing point, consistent scale recession, rich detail, and | Anthropic: Claude Fable 5.1: Model A delivers a clean, symmetric one-po…
Matchup powered by OpenRouter.