Krea 2 Medium Review: The Line's Least Filtered Tier

The gallery holds published, safe-for-work examples from the Krea 2 Medium model page. Our content-ladder frames aren't shown; we report those results in prose instead.

One candle, one dancer. A ballet dancer at a wooden barre lit only by a single candle flame, the rest of the room falling into a Dutch-master's near-black — the frame this page opens on.

Steam as the subject. A woman in a cedar bathtub with a pale fox mask floating in the rising steam behind her, the whole composition built around one horizontal waterline at the collarbone.

A door left ajar. An available-light shot seen through a half-open hotel-room door: a woman in an oversized sweater on the edge of a made bed, absorbed in her phone, lit by one bedside lamp.

Grain and downpour. A black-and-white 35mm frame of a fashion model in a soaked slip dress, arms flung wide in a torrential downpour on an empty crosswalk at night.

Krea 2 Medium takes the same wording style as the rest of the line: name the film stock, the single light source, and the material, then let the rest of the 5,000-character budget do the atmosphere. It responds better to one clear light decision than to a long list of qualities.

1. One light source, named directly. The prompt behind the candlelit ballet frame above.

Chiaroscuro oil-painting light, a single candle flame as the only source, Dutch master palette of deep umber and near-black with one warm amber pool of light, heavy falloff, visible dust motes. An adult ballet dancer stands at a wooden barre, one leg extended, shoulders and collarbone catching the only warm light in the frame.

Run this prompt on Krea 2 Medium →

2. A waterline as the compositional anchor. The prompt behind the bathhouse frame: everything above and below one horizontal line is treated differently.

Cinematic film still, old Japanese bathhouse. A woman lies back in a cedar tub with her head tipped against the rim and eyes closed, the waterline at her collarbone, wet hair plastered to her shoulder; a pale fox mask floats in the steam behind her, rising vapour catching warm lantern light.

Run this prompt on Krea 2 Medium →

3. Grain and a single weather condition, named plainly. The prompt behind the rain-crosswalk frame.

Black-and-white 35mm photograph, pushed grain, full-length. A tall adult fashion model stands in the middle of an empty city crosswalk in a torrential downpour at night, wearing a soaked black satin slip dress and heels, arms flung wide, head tipped back into the rain.

Run this prompt on Krea 2 Medium →

Weak: "Woman in a bathtub, relaxing." No light source, no single visual anchor. Strong: prompt 2 — one waterline, one light, one prop doing the mood.

Krea 2 Medium reached further up our content ladder than either sibling tier. Across 16–25 September 2026 runs, L1 Suggestive rendered every time; L2 Mature themes was refused every time, which is the one rung out of order on this tier; L3 Strong rendered in two of four tries and came back softened in the other two; L4 Intense, our L5 Extreme restraint-and-impact probe, and our L5 Extreme gore probe all came back softened rather than refused. The page label is Softened: expect L1–L3 content to come through at the lower rungs, with upper-rung content toned down rather than blocked outright.

Not tested yet: whether the L2 Mature themes refusal is consistent across more tries, since our sample there was small; anime and illustrated styles on the ladder; scenes with two women or two men above L1 Suggestive.

Adults only, in every prompt. Never depict a real, identifiable person without their consent.

The Krea 2 line's most pointed criticism from outside is about how the base model was tuned. One commenter, discussing what Krea's own team calls "alignment training," argued it functions as "a code word for censorship/lobotomy" — a claim our own results complicate, since Medium reached L3 Strong more often than its siblings even as one specific rung, L2, was refused outright every time.

On image quality, a recurring complaint is softness: one commenter described output from this part of the line as "the same blurry, airbrushed output" as another model sharing a similar image decoder. Our own portrait test showed less skin texture than the brief called for, which lines up with that complaint, even as sharper single-light scenes like the candlelit barre came through with real material detail.

• Same line, more finishing detail? Krea 2 Large adds anamorphic ratios and reference images, at $0.06 an image, but refused nearly everything on our content ladder. Our Krea 2 Large review.
• Same line, built for speed? Krea 2 Medium Turbo adds anamorphic ratios and reference images too, at $0.015. Our Krea 2 Medium Turbo review.
• Sharper skin and a working editor? Seedream 5.0 Pro passed our upper content-ladder rungs from text and as an edit, from $0.036 at 1K. Our Seedream 5.0 Pro review.
• A cheap, fast base with a seed? Z-Image Turbo runs $0.01 an image with repeatable results across a series.