Head to head: Ernie Image Lora vs Luma Uni-1 Edit Max

Luma Uni-1 Edit Max takes this comfortably and decisively. The aggregate gap is large — 69.7 to 56.5 — and the statistical read is even clearer: a 99% confidence win. More telling than the raw score, though, is the task ledger: Luma wins 6 tasks, Ernie wins 0, with 2 ties. That is not a stylistic preference split; it is one model repeatedly executing the brief better.

The pattern is straightforward: Luma was the more obedient model. It won on perspective and scale with the empty one-point-perspective library aisle, on negation by actually excluding forbidden objects, on exact counting with a true overhead seven-cup flat lay, and on the Marshgate shelf tag by delivering the requested Swiss-style placard and line breaks instead of improvising. It also handled the rainy citrus scooter ad better, getting closer to the requested compact electric scooter, branding, and flat-vector treatment. These are not tiny aesthetic quibbles — they are the kinds of prompt-following failures that decide whether an image is usable.

Ernie Image Lora’s problem is that it often looked good while being wrong. The library scene was attractive but sabotaged by the large foreground book/table that broke the empty-aisle brief. The cozy reading nook ignored negation constraints with visible wall art and shelving. The shelf-tag task drifted into a supermarket scene and collapsed the typography. Even where it nearly hung in, it had a habit of slipping off the exact ask: right count but wrong camera angle, nice mood but weaker context, polished composition but missing the specified object form.

The two ties are revealing. In hands & anatomy, the judges split on which image better conveyed the bracelet-tying action, suggesting neither model fully locked the assignment. In missed-train commuter, each model captured different parts of the brief — Ernie stronger on narrative props and context, Luma stronger on subtle facial emotion and cinematic portraiture. Those are fair draws. The kettle task also shows Luma’s win was not effortless: one pass preferred Ernie’s simpler stovetop product read, the other preferred Luma’s cleaner palette and studio restraint. But across the full set, Luma was still the model with fewer unforced errors.

Final call: Luma Uni-1 Edit Max is the clear winner. Ernie Image Lora has moments of visual appeal, but Luma is the model you trust when the prompt actually matters.

How they were tested

We ran 8 fresh image tasks, generated on the fly for this matchup so neither model could prepare in advance, and had gpt-5.4 score each one. To cancel position bias, every task was judged twice — once in each presentation order — and every number reported here, including the headline totals, is the average of both passes. Ernie Image Lora scored 56.5 to Luma Uni-1 Edit Max's 69.7.

1. 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.

Winner: Luma Uni-1 Edit Max — Model B matches the prompt much more closely with a clean one-point-perspective view down a long empty library aisle, consistent scale recession, and warm overhead lighting. Model A is visually appealing but breaks prompt adherence by placing a large foreground book/table in the aisle and weakening the intended empty-aisle perspective. (Second judge pass, order swapped — scores are the average of both: Model B matches the prompt much more closely with a photorealistic, empty library aisle in clear one-point perspective, consistent shelf recession, and warm overhead lighting. Model A has pleasing symmetry, but it violates the prompt by placing a large book on a table in the aisle and feels less like a long empty aisle with a clean vanishing-point view.)

2. 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.

Winner: Tie — Model B better matches the prompt by showing a person actively tying bracelet threads with both hands clearly visible and anatomically plausible finger positions. Model A is attractive and well lit, but it reads more like holding or presenting a finished bracelet than candidly tying one, and the extra blurred hand in the background weakens prompt adherence. (Second judge pass, order swapped — scores are the average of both: Model A better matches the prompt with both hands clearly visible, a more plausible bracelet-tying action, and stronger shallow-depth-of-field candid photography. Model B has decent realism and anatomy, but the grip and thread interaction are less convincing and the scene feels more like handling loose string than tying a friendship bracelet.)

3. Rainy citrus scooter ad

A clean stylized vector illustration of a compact electric scooter branded "LumaZip S4" parked beside a corner fruit stand at dusk in light rain, with glossy tangerines stacked in crates and reflections on wet pavement; modern flat-shape illustration with crisp outlines, subtle grain, and a cheerful-but-lonely mood, viewed in three-quarter angle with the scooter centered and warm shop light contrasting cool blue evening tones, 16:9.

Winner: Luma Uni-1 Edit Max — Model B adheres more closely to the prompt with a compact electric scooter shown in a clear three-quarter view, visible branding close to "LumaZip S4," stronger flat-vector styling, and a better balance of warm shop light against cool rainy dusk tones. Model A is attractive and centered with nice wet reflections, but it looks more like a kick scooter than the advertised compact scooter and lacks the requested branding/text detail. (Second judge pass, order swapped — scores are the average of both: Model B matches the prompt more closely with a compact scooter centered in three-quarter view, visible rainy dusk atmosphere, warm fruit-stand lighting against cool blue tones, wet-pavement reflections, and readable branding close to “LumaZip S4.” Model A is attractive and polished, but it depicts a kick scooter rather than the requested compact electric scooter, lacks the specified branding text, and feels less aligned with the stylized flat-vector brief.)

4. 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.

Winner: Luma Uni-1 Edit Max — Model B adheres much better to the negation constraints, showing the armchair, books, mug, and window without any visible plants, lamps, or artwork. Model A is cozy and well-made, but it clearly includes framed wall art and a large bookshelf, making it less faithful to the prompt despite strong aesthetics. (Second judge pass, order swapped — scores are the average of both: Model B adheres closely to the prompt with a cozy armchair, books, tea by a window, warm afternoon light, and importantly no visible plants, lamps, or artwork. Model A is attractive but violates the negation constraints with visible framed artwork and a large bookshelf dominating the scene, making it less faithful despite strong aesthetics.)

5. Fjord enamel palette kettle

A studio product image of a stovetop kettle called "Northwake Mini" rendered using only the named Fjord Enamel palette: glacier white, deep navy, seafoam green, muted salmon, and fog gray, with absolutely no other colors anywhere in the image; soft diffused tabletop lighting, simple side profile, smooth enamel finish, minimal shadow, restrained modern product-photography composition, 16:9.

Winner: Luma Uni-1 Edit Max — Model A better matches the requested studio product photo of a stovetop kettle with a simple side-profile composition, soft lighting, and mostly correct palette use; its main flaw is the small dark metal details and slightly imperfect text rendering. Model B is elegant and clean, but it departs more from the prompt by adding large promotional text in the scene, using a less clearly stovetop-kettle form, and introducing gray elements more prominently in a layout that feels more like an ad than restrained product photography. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested restrained modern studio product shot with a simple side-profile composition, soft lighting, smooth enamel finish, and clean text rendering; it also stays closer to the specified palette overall. Model A is visually appealing but deviates more from the prompt with a less side-on angle, a more traditional kettle form, darker metallic details, and less controlled adherence to the palette-only constraint.)

6. Missed-train commuter

A close-up cinematic portrait of a commuter in their early 30s clutching a reusable coffee tumbler and a folded grocery receipt, standing on a nearly empty suburban platform just as the train tail lights disappear behind them; the face must convincingly show the specific emotion of stunned frustration shifting into disbelief, with tense jaw, raised inner brows, parted lips, and glassy eyes, natural skin texture, realistic photography, cool dawn light with shallow depth of field.

Winner: Tie — Model A better matches the prompt with a clearer missed-train context, visible disappearing train, reusable tumbler and receipt, and a stronger expression of stunned frustration shifting into disbelief. Model B has a convincing portrait and mood, but the train context is weaker, the platform reads darker than cool dawn, and the emotion is more subdued and less specific. (Second judge pass, order swapped — scores are the average of both: Model B better matches the prompt’s close-up cinematic portrait and the nuanced emotion of stunned frustration turning to disbelief, with convincing glassy eyes, tense jaw, raised inner brows, and realistic dawn-platform atmosphere; the disappearing train tail lights in the background also align well. Model A is strong visually, but the train appears to be arriving rather than already gone, the expression reads more exaggerated panic than disbelief, and the receipt/text handling is less convincing.)

7. Marshgate shelf tag

A sharply legible retail shelf sign for a sparkling water display, designed as a minimalist in-store placard reading exactly: "MARSHGATE FIZZ\nYuzu Lime\n8 PACK\n$5.49"; clean Swiss-style typography, black text on matte cream card with a thin teal border, straight-on composition, even supermarket lighting, no extra words, no decorative clutter, text perfectly readable, 16:9.

Winner: Luma Uni-1 Edit Max — Model B adheres more closely to the requested Swiss-style placard with the exact four-line text, clean centered layout, matte cream card, and thin teal border. Model A is readable and attractive, but it collapses multiple lines into one, adds a supermarket background that distracts from the minimalist placard, and is less faithful to the specified sign design. (Second judge pass, order swapped — scores are the average of both: Model B adheres more closely to the prompt with the exact four-line text layout, minimalist Swiss-style treatment, matte cream card, thin teal border, and no extra environmental clutter. Model A is attractive and readable, but it places the sign in a supermarket scene and collapses three lines into one, reducing fidelity to the requested straight-on placard design.)

8. 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.

Winner: Luma Uni-1 Edit Max — Model B adheres more closely to the prompt by presenting an actual overhead flat lay on white marble with exactly seven clearly separated identical espresso cups in a straight row. Model A gets the count and row right, but it is not overhead and reads more as a side/front view, so Model B also feels cleaner and more compositionally aligned. (Second judge pass, order swapped — scores are the average of both: Model B matches the prompt much more closely: it shows exactly seven identical espresso cups in a clearly countable single straight row on white marble with an overhead flat-lay perspective. Model A also has seven cups and a clean row, but it misses the required overhead flat-lay viewpoint and is less exact in prompt adherence despite strong aesthetics and technical polish.)

See every prompt and the full side-by-side outputs in the interactive Head-to-Head.