Head to head: FLUX 2 vs Ideogram V4.5 Edit
FLUX 2 vs Ideogram V4.5 Edit
By Ryan Merket · Published
A varied image-model comparison spanning exact object counts, material and spatial relationships, macro detail, hands, and tightly constrained scenes. The verdict weighs both prompt fidelity and how cleanly each image communicates its details.
FLUX 2’s aggregate score was **63.7**, compared with **58.8** for Ideogram V4.5 Edit. Yet task-level verdicts split evenly: FLUX 2 took exact counting, attribute binding, cryogenic textile micro-detail, and the isotope hall; Ideogram took the pollen assay, hands and anatomy, negation, and the spectrometry desk. The saved result is a tie, not an overall win for either model. FLUX 2’s strongest showings were cases where visual discipline mattered: its seven espresso cups were more uniform and cleanly spaced, its textile close-up surfaced fine fibers and frost, and its isotope hall held to a coherent grayscale perspective. Ideogram was more persuasive in several scenes that benefited from a polished, legible setup—especially the macro assay and the spectrometry kit—though it sometimes introduced distractions or departures from the brief. Neither model swept a consistent category. Ideogram’s craft-table scene looked more natural, while FLUX 2 made the requested hand-thread interaction clearer; the cozy nook was appealing in both, with different compositional compromises. Those mixed results make the higher aggregate score insufficient grounds to overturn the saved tie. **Final call: Too close to call (not statistical). This assessment applies only to the saved task sample.**
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 A clearly shows exactly seven identical espresso cups in a clean, straight row on white marble, while Model B has seven cups but adds saucers and inconsistent, mostly empty contents. Model A also has more even spacing and a cleaner, more cohesive presentation. (Second judge pass, order swapped — scores are the average of both: Model A presents exactly seven clearly separate, nearly identical espresso cups in a straight overhead row on white marble. Model B also has seven cups, but the sauc | Anthropic: Claude Fable 5.1: Both Model A and Model B render exactly seven c…
Macro pollen assay frame
A macro close-up of a custom glass microfluidic slide in a botany lab, viewed at a low angle through a 100mm macro lens so the nearest etched channels and amber pollen grains are tack-sharp while the background dissolves into soft optical bokeh; include a gloved fingertip just outside the focal plane, iridescent interference colors along the glass edge, cinematic scientific photography, narrow depth of field, 16:9.
OpenAI: GPT-6 Luna: Model B presents the etched channels and pollen in a particularly clear, polished macro composition, with strong depth and iridescence along the glass edge. Model A also follows the prompt closely and has more pronounced bokeh, while Model B’s extra specimen label is a minor unrequested addition. (Second judge pass, order swapped — scores are the average of both: Model B more fully conveys the low-angle scientific macro scene, with crisp etched channels and amber pollen, pleasing bokeh, and a vi | Anthropic: Claude Fable 5.1: Model A nails the macro brief with genuinely ta…
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 clearly matches the requested object arrangement and material contrast, with a strong, balanced tabletop composition. Model B includes the correct objects and placements but leaves substantial empty space and renders the cube noticeably pale and less distinct as green. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested object relationships and finishes more closely, especially with a matte green cube and glossy red sphere. Model A has a | Anthropic: Claude Fable 5.1: Model A nails every binding—matte green cube le…
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 A clearly shows both hands gripping the threads, but the fingers look somewhat awkward and the close crop feels less candid. Model B has a more natural, polished craft-table scene with convincing hand anatomy, though one hand is hovering rather than visibly gripping the threads. (Second judge pass, order swapped — scores are the average of both: Model B more clearly depicts a candid bracelet-making interaction, with both hands plausibly working the colored threads and a natural tabletop se | Anthropic: Claude Fable 5.1: Model A shows both hands very clearly with most…
Cryogenic textile micro-detail
A hyperrealistic materials-science lab scene showing a cryogenic technician’s cobalt-blue insulated glove resting on a brushed titanium workbench beside a frost-rimmed vial rack, with the glove’s dense fabric weave, stray fibers, stitched seams, ice crystals, hairline scratches in the metal, and tiny engraved serial marks on a micrometer rendered razor-sharp and unsmudged; cool white cleanroom lighting, crisp directional highlights, documentary realism, 16:9.
OpenAI: GPT-6 Luna: Model A better captures the requested micro-detail, with a sharply rendered glove weave, stray fibers, stitched seams, frost, and scratched titanium. Model B includes more lab context and a visible micrometer, but its details are softer and the extra glove intruding from the right distracts from the scene. (Second judge pass, order swapped — scores are the average of both: Model A better captures the requested micro-detail, with sharply visible glove weave, loose fibers, stitched seams, frost, a | Anthropic: Claude Fable 5.1: Model A delivers the requested micro-detail: de…
Vanishing-point isotope hall
A wide interior view of an experimental isotope-imaging facility in minimalist line art with subtle grayscale shading, looking down a long corridor of evenly spaced shielded scanner bays toward a single clear vanishing point; in the foreground place a rolling lead-lined sample cart, a waist-high technician, and a towering ceiling gantry crane with all relative sizes physically believable, parallel architecture converging consistently, floor tiles diminishing correctly with distance, and no warped walls or mismatched scale, 16:9.
OpenAI: GPT-6 Luna: Model A more clearly fulfills the restrained grayscale line-art brief, with a strongly centered vanishing point, evenly spaced shielded bays, and a convincing lead-lined cart beneath a coherent gantry. Model B is polished and detailed, but the bright yellow warning sign and garbled signage depart from the requested subtle grayscale treatment. (Second judge pass, order swapped — scores are the average of both: Model A better conveys a full ceiling gantry crane and a consistently receding, orderly | Anthropic: Claude Fable 5.1: Model A delivers heavy shielded scanner bays, a…
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 creates a particularly warm, inviting nook and clearly includes the chair, books, tea, and window while avoiding all prohibited items. Model B also follows the constraints, but its wide blank wall and books on the floor make the scene feel less cohesive and cozy. (Second judge pass, order swapped — scores are the average of both: Model B includes the requested nook elements and warm window light while keeping plants, lamps, and artwork out of view. Model A is appealing and intimate, but | Anthropic: Claude Fable 5.1: Both Model A and Model B correctly omit plants,…
Zorn-palette spectrometry desk
A top-down flat lay illustration of a compact field spectrometry kit arranged with meticulous spacing on a matte drafting surface: foldable spectrometer, calibration card, lab notebook, hex key, sample vials, and a handwritten label reading 'Site J-47'; render the entire image strictly in the named Zorn palette only—ivory white, yellow ochre, vermilion red, and ivory black—with no blue, green, purple, or stray off-palette hues anywhere; minimalist editorial realism with controlled soft studio lighting, 16:9.
OpenAI: GPT-6 Luna: Model B more clearly conveys a field spectrometry kit, with a detailed foldable instrument, readable site label, and calibration card, while staying close to the requested palette. Model A has cleaner spacing and a clearly recognizable hex key, but its instrument and calibration card are less specific and detailed. (Second judge pass, order swapped — scores are the average of both: Model B better depicts a foldable spectrometer and a clearly rendered calibration card while staying close to the s | Anthropic: Claude Fable 5.1: Model A honors the strict top-down flat lay and…
Matchup powered by OpenRouter.