Head to head: Flux 3 Image vs Wan
Flux 3 Image vs Wan
By Ryan Merket · Published
A prompt-by-prompt image-model comparison tests style fidelity, object placement, exact counts, emotional expression, reflections, and palette control. The results show how each model handles the difference between making an attractive image and following the brief precisely.
Flux 3 Image wins this matchup decisively: it takes seven of eight tasks, leads 64.2 to 51.8 overall, and the evaluation puts the result at limited confidence. The gap comes less from raw polish than from reliably honoring specific instructions. That shows up across very different briefs. Flux delivered the more faithful ukiyo-e fishing scene, put the apple on the counter rather than chiefly inside the glass, and followed the wheelbarrow staging more closely. It also hit exact-count requirements: three courtyard volunteers and seven visible terracotta pots, with the requested details easier to distinguish. Flux was stronger on subtler instructions, too. Its orchid keeper reads as relieved rather than simply sad, and its reading nook avoids the explicitly forbidden plants, lamps, and artwork. Wan’s visible plants outside the window made that negation miss especially consequential. These are not glamorous wins; they’re the basic prompt-following details that make an image usable. Wan’s win is real and specific: it better respected the restricted four-color, flat-vector brief, while Flux’s gradients and tonal variation broke the constraint. But one well-targeted advantage could not offset Flux’s consistency across the other tasks. **Verdict: Flux 3 Image wins, decisively—especially when the brief has details that must not be improvised away.**
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-6 Luna: Model A more fully realizes the ukiyo-e woodblock style with bold outlines, stylized foam, a muted indigo-and-cream palette, and a clearly occupied fishing boat riding the cresting wave at dawn. Model B has a clean, appealing composition, but its boat is less distinct and lacks the fishermen that make Model A’s scene more faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model A more clearly depicts a fishing boat with fishermen riding the wave, while re | Anthropic: Claude Fable 5.1: Model A delivers an authentic woodblock texture…
Reflections & glass
A photorealistic close-up of a chrome teapot on a polished dark marble counter next to a glass of water; the window and a red apple on the counter must be correctly reflected in BOTH the chrome and the water's surface, dramatic side light, 16:9.
OpenAI: GPT-6 Luna: Model A clearly shows the red apple on the counter and convincing window reflections in the chrome, while Model B’s apple appears mainly inside the glass rather than on the counter. Both have attractive polished surfaces, but Model A better fulfills the requested scene and reflection setup. (Second judge pass, order swapped — scores are the average of both: Model A includes the apple on the counter and gives a convincing chrome reflection of the window and apple, though the apple’s reflection in | Anthropic: Claude Fable 5.1: Model A places an actual red apple on the count…
Wheelbarrow staging puzzle
Hyperreal editorial garden scene, 16:9 — in a tidy potting shed doorway, a red wheelbarrow is centered; a stack of three seed trays sits on top of the wheelbarrow, a coil of green twine lies to the left of the wheelbarrow, a copper watering can is to the right of the wheelbarrow, a muddy pair of rubber boots is behind the wheelbarrow, and a small crate of orange marigold seedlings is precisely between the twine and the watering can in the foreground; late afternoon sun slanting in from the right, balanced symmetrical composition, everything clearly separated and visible.
OpenAI: GPT-6 Luna: Model A follows the requested staging especially closely, with the wheelbarrow centered, boots behind it, and the twine, marigold crate, and watering can clearly arranged across the foreground. Model B includes the key objects but uses a side-on wheelbarrow and a less symmetrical layout, making the staging less precise. (Second judge pass, order swapped — scores are the average of both: Model B more closely captures the warm late-afternoon light from the right, and its seedling-filled trays and | Anthropic: Claude Fable 5.1: Model A places every object as specified—trays o…
No-gnome courtyard
Documentary photojournalism, 16:9 — a rain-damp urban courtyard converted into a community herb garden at blue hour, with waist-high galvanized planters overflowing with basil, dill, and purple shiso, a folding table covered in seed packets, a yellow hose coiled near a brick wall, and three volunteers in muddy aprons pruning and labeling pots; the scene must contain absolutely no garden gnomes or other decorative statues anywhere in the frame; soft sodium-vapor streetlight mixed with cool twilight, natural candid composition from eye level.
OpenAI: GPT-6 Luna: Model A closely matches the requested candid, rain-damp blue-hour courtyard, including exactly three volunteers, labeled herbs, galvanized planters, seed supplies, and a coiled yellow hose. Model B captures much of the setting and lighting but shows four volunteers and feels more staged in its composition. (Second judge pass, order swapped — scores are the average of both: Model A matches the requested three volunteers and clearly shows labeled herbs, seed-starting materials, wet paving, and the | Anthropic: Claude Fable 5.1: Model A delivers exactly three distinct volunte…
Seven terracotta cuttings
A crisp studio product photograph, 16:9 — exactly seven distinct terracotta pots, each holding a different rooted cutting of a houseplant (monstera, peperomia, pilea, tradescantia, pothos, begonia, and rosemary), arranged in a gentle arc on a matte sage-green tabletop, every pot fully visible and individually countable, with small handwritten plant tags tucked into the soil; clean diffused morning light from the left, shallow shadows, high-detail realistic texture.
OpenAI: GPT-6 Luna: Model A clearly presents exactly seven fully visible terracotta pots in a gentle arc, with distinct plants and mostly legible handwritten tags against a sage-green setting. Model B also has seven pots and an appealing arrangement, but its tags are less readable and the plant varieties are less clearly differentiated. (Second judge pass, order swapped — scores are the average of both: Model A shows exactly seven distinct, fully visible terracotta pots with recognizable examples of the requested p | Anthropic: Claude Fable 5.1: Model A delivers all seven requested species id…
Relieved orchid keeper
Cinematic close portrait, 16:9 — a middle-aged orchid collector in a greenhouse, holding a newly revived pale apricot orchid bloom near her chest with both hands, convincingly expressing intense relief after fearing the plant had died: softened eyes with visible moisture, eyebrows gently raised inward, mouth parted in a small exhale that reads as release rather than joy, shoulders dropping; rows of blurred hanging orchids behind her, humid air catching warm sunrise light through glass panes, shallow depth of field, realistic skin texture and natural color.
OpenAI: GPT-6 Luna: Model A conveys relief more convincingly through softened, moist eyes and a small exhale-like expression, while also showing the greenhouse and sunrise setting. Model B has a strong close portrait, but the expression reads more as grief or distress than release. (Second judge pass, order swapped — scores are the average of both: Model B has the tighter portrait and more visibly tearful emotion, but its expression leans toward sadness and the bloom appears disconnected from a stem. Model A more c | Anthropic: Claude Fable 5.1: Model A delivers a richer cinematic greenhouse…
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 more closely follows the flat-vector, four-color restriction, while Model A uses noticeable gradients, texture, and extra tonal variation. Model A has a more immersive, cohesive campsite composition, but Model B’s clear graphic treatment better satisfies the defining constraint. (Second judge pass, order swapped — scores are the average of both: Model B adheres more closely to the four-color, flat-vector restriction, while Model A has stronger scenic polish but uses noticeable tonal vari | Anthropic: Claude Fable 5.1: Model A is the more atmospheric and polished sc…
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 clearly includes the reading nook essentials and avoids indoor plants, lamps, and artwork; Model B has visible potted plants outside the window, violating the explicit negation. Both are warm, polished compositions, though Model A follows the constraints more closely. (Second judge pass, order swapped — scores are the average of both: Model B includes visible plants outside the window, directly violating the prompt’s explicit exclusion. Model A includes the requested reading-nook element | Anthropic: Claude Fable 5.1: Model A delivers the armchair, book stack, stea…
Matchup powered by OpenRouter.