Head to head: Hidream I1 Dev vs Ideogram V4.5 Edit
Hidream I1 Dev vs Ideogram V4.5 Edit
By Ryan Merket · Published
A constraint-heavy test of two image models with sharply different strengths, spanning exact counts, typography, cinematic composition, perspective, and complex narrative scenes.
Ideogram V4.5 Edit posts the higher aggregate score, 57.8 to 51.7, but that six-point gap is not decisive: the analysis assigns only limited confidence that either model is genuinely better. In practical terms, this is a sample tie—not a narrow win dressed up as certainty. Ideogram was more dependable on rigid constraints. It repeatedly beat Hidream on the seven-cup counting test, came closer to the requested restricted-palette vector style, and captured more of the specified macaque actions and crate text. But its habitual square framing cost it prompt adherence, and its advantage did not hold consistently on typography or the nine-lanternfish task. Hidream was stronger when the brief demanded a widescreen editorial composition. It consistently won the Heron Sketcher Hands scene and delivered the more convincing 16:9 noir treatment for the terrarium setup, while the perspective tests ended level: Hidream respected framing and warmth, whereas Ideogram produced more photorealistic depth. The split results on text, reflections, and specimen counting expose meaningful run-to-run fragility in both models. **Final call: tie—too close to call. Choose Ideogram V4.5 Edit for counting, constrained graphics, and detail-heavy prompt coverage; choose Hidream I1 Dev for cinematic 16:9 composition and environmental storytelling. Neither establishes an overall edge here.**
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-5.6 Sol Pro: Model B correctly shows exactly seven clearly separate espresso cups in one straight row on white marble, though its square framing misses the requested 16:9 aspect ratio. Model A uses a suitable overhead style and 16:9 frame but shows ten cups arranged in two rows, violating the core exact-counting requirement. (Second judge pass, order swapped — scores are the average of both: Model B clearly shows exactly seven separate cups in one straight row on marble, though it misses the requested 16:9 a | Anthropic: Claude Fable 5.1: Model A shows ten cups in two rows of five…
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 A closely follows the requested warm lighting, empty one-point-perspective aisle, consistent recession, and 16:9 framing, though its highly uniform books look slightly synthetic. Model B has stronger photorealism and an especially convincing sense of depth, but its square format directly misses the specified aspect ratio and some foreground elements appear mildly oversized or distorted. (Second judge pass, order swapped — scores are the average of both: Model B is substantially more photor | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 frame…
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-5.6 Sol Pro: Model B better captures the flat-vector, restricted-palette intent with cleaner color regions, though Model B is square rather than 16:9 and still uses multiple orange and teal shades. Model A has a stronger widescreen composition but clearly introduces gradients, texture, and numerous off-palette tonal variations. (Second judge pass, order swapped — scores are the average of both: Model B better captures the requested flat-vector, largely gradient-free treatment, though it uses more than four s | Anthropic: Claude Fable 5.1: Model A delivers the required 16:9 frame w…
Rainhouse Terrarium Reflections
A hyperreal cinematic interior of a remote rainforest research station at night, centered on a large glass terrarium wall containing orchids, moss, and a red-eyed tree frog, with a chrome thermos, a shallow puddle on the concrete floor, and a round inspection mirror hanging nearby; the scene must show physically plausible reflections and transparency in all surfaces, including correct mirrored room details, subtle refraction through the glass, water reflections from the frog habitat, and soft glare from a single green-shaded desk lamp. Moody film-noir-inspired lighting in muted color, dense humidity, precise perspective, 16:9.
OpenAI: GPT-5.6 Sol Pro: Model A better fulfills the requested 16:9 cinematic film-noir presentation, with stronger composition, atmosphere, and integration of the lamp, thermos, glass, mirror, and puddle, though the frog is incorrectly outside the terrarium. Model B places the frog correctly within the habitat and has convincing puddle reflections, but its square format, oversized frog, flatter lighting, and less informative mirror reflection reduce adherence and realism. (Second judge pass, order swapped — scores are | Anthropic: Claude Fable 5.1: Model A delivers the required 16:9 frame, m…
Heron Sketcher Hands
A realistic editorial wildlife portrait of a field ornithologist kneeling on a salt-marsh boardwalk at dawn, clearly visible from head to knees, holding a graphite sketchbook in her left hand while her right hand gently adjusts the focus ring on a compact spotting scope; both hands are fully shown with all five fingers visible and natural, accurate anatomy, believable wrists and elbows, and correct body proportions, with a great blue heron standing beside her in the reeds. Soft peach sunrise rim light, crisp coastal mist, detailed outdoor clothing, natural pose, shallow depth of field, cinematic composition, 16:9.
OpenAI: GPT-5.6 Sol Pro: Model A better delivers the requested 16:9 cinematic portrait, kneeling pose, full head-to-knees framing, boardwalk, sunrise, and prominent heron, though the handheld optic is not a proper spotting scope and all fingers are not clearly visible. Model B renders the hands and focus-ring interaction more convincingly, but its square, tightly cropped composition omits the full figure and weakens the editorial environmental portrait. (Second judge pass, order swapped — scores are the average of both: | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 cinema…
Nine Lanternfish Specimens
A meticulous natural-history studio still life in luminous color realism: exactly 9 distinct preserved lanternfish specimens arranged on a matte black examination tray, each separated from the others and individually countable, each with a tiny handwritten ivory tag tied to its tail, plus scattered droplets of seawater and a brass caliper near the tray edge. Overhead diffused museum lighting, ultra-sharp focus, dark background, symmetrical composition, no extra fish, no overlapping bodies, 16:9.
OpenAI: GPT-5.6 Sol Pro: Model B is more convincing as a natural-history specimen arrangement, with distinct separated fish, droplets, and a brass caliper, though it has 10 fish rather than 9, incomplete tagging, a blue tray, and a square aspect ratio. Model A is attractively symmetrical and correctly widescreen but shows only 7 near-identical fish, improperly positioned tags, and an unrequested stopwatch. (Second judge pass, order swapped — scores are the average of both: Model B is more naturalistic, varied, and visua | Anthropic: Claude Fable 5.1: Model B correctly delivers exactly nine in…
Legible multi-line text
A minimalist event poster with three lines of crisp, correctly-spelled text stacked and centered: 'NIGHT MARKET' large on top, 'Fridays · 6–11pm' in the middle, 'Riverside Pier 4' at the bottom, on a deep navy background, clean sans-serif, subtle grain.
OpenAI: GPT-5.6 Sol Pro: Model B closely matches the deep navy, centered minimalist poster brief with crisp, fully legible text and a tasteful grain treatment, though the title wraps and the capitalization differs from the requested wording. Model A has a less faithful split-color layout and renders the time incorrectly as “6--11pm,” weakening both adherence and polish. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the deep navy, minimalist, centered poster aesthetic with cr | Anthropic: Claude Fable 5.1: Model B delivers all three lines correctly…
Macaques and Tea Crates
A richly detailed storybook-gouache scene of 6 Formosan macaques working together in a misty mountain tea yard: two tugging a bamboo rope, one perched on a weathered crate, one sniffing spilled tea leaves, one reaching toward a dangling copper bell, and one juvenile peeking from under a handcart, all interacting naturally around stacked cedar tea chests stamped with the fictional mark 'Lanyue Ridge Cooperative'. Balanced multi-subject composition, layered depth, cool monsoon light with warm lantern accents, coherent shared ground plane, lush ferns and stone steps, 16:9.
OpenAI: GPT-5.6 Sol Pro: Model B includes all six macaques, more of the specified individual actions, lush misty depth, and substantially more accurate cooperative text, though it misses the 16:9 format and copper-bell interaction. Model A has a cleaner widescreen composition and strong technical polish, but appears to show only five macaques, replaces the requested text with symbols, and omits several key actions. (Second judge pass, order swapped — scores are the average of both: Model B includes all six macaques, sev | Anthropic: Claude Fable 5.1: Model A nails the storybook-gouache look,…
Matchup powered by OpenRouter.