Krea 2 Large is #2 for style fidelity

The Krea Team May 12, 2026

Contra Labs ran Krea 2 Large through a four-model style-transfer evaluation against GPT Image 2, Gemini 3 Pro Image Preview, and Seedream 5.0 Lite. Their headline: Krea 2 Large is the #2 style-transfer model, with a reported Style Fidelity gap of 0.14 points to GPT Image 2.

The test is especially meaningful because it isolates the thing Krea 2 was built for: holding onto a reference image's visual language. Color palette, brushwork, tonal quality, texture, and the overall direction of the style all matter more here than simply making the most immediately dramatic image.

You can read the full Contra Labs study here, or see Ben's original thread on X.

Contra Labs announcement graphic: Creatives vote, Krea 2 is number two for style fidelity
Source: Contra Labs style-transfer evaluation, May 2026.

Style fidelity

On Style Fidelity, Krea 2 Large sits close to GPT Image 2 and well ahead of the rest of the field. Gemini 3 Pro Image Preview and Seedream 5.0 Lite both fall below the neutral midpoint on this dimension.

That puts Krea 2 Large above the 3.0 line where outputs move from net-negative to useful, and makes it the only non-GPT model in the evaluation to clear that threshold on style transfer.

Average scalar rating by model across prompt adherence, visual quality, usability, and style fidelity
Krea 2 Large narrows the field most clearly on Style Fidelity.
Style transfer comparison showing Krea 2 Large preserving an illustration style while other models drift toward photorealism
Contra Labs highlighted Krea 2 Large as the only model in this example that did not drift to photorealism.

Ceiling and floor

The top of the rating scale is where the result gets interesting. Krea 2 Large received a 5 on Style Fidelity for 23% of outputs. GPT Image 2 received a 5 for 22%.

At the peak, Krea 2 Large is already operating at the same rate as the frontier model. The difference shows up in the floor: 26% of Krea's outputs landed in the bottom two ratings, compared with 20% for GPT Image 2 and 46% for Gemini 3 Pro Image Preview.

Rating distribution for Style Fidelity showing Krea 2 Large at 23 percent 5 ratings and 26 percent bottom-two ratings
When Krea 2 Large hits, it hits at the frontier rate. The floor is where GPT Image 2 still leads.

What it means

Krea 2 Large does not win every format of the evaluation. In head-to-head voting, it placed fourth overall at 39.5% and lost pairwise matchups to Seedream and Gemini, even though those models scored lower on the absolute style-fidelity rubric.

That tension is the point. Pairwise tournaments reward images that feel immediately compelling. Style-transfer briefs reward restraint: applying a reference to a new subject without drifting away from the reference. Krea 2 Large is strongest when the reference is the direction, not just the starting point.

Strengths and weaknesses chart for Krea 2 Large, with style fidelity as the clearest strength
Evaluators called out style fidelity as Krea 2 Large's clearest strength.
Head-to-head win rate matrix for GPT Image 2, Gemini 3 Pro Image Preview, Seedream 5.0 Lite, and Krea 2 Large
H2H voting rewards the most compelling image in the set, which is a different signal from absolute style fidelity.

Use it for style briefs

The practical read is narrow and useful: use Krea 2 Large when the brief is a style brief. If the reference image carries the visual direction and the job is to transfer that style to a new subject without drift, Krea 2 Large is already competitive near the top of the field.

For prompt-led work, where the text brief carries more of the creative load than the reference, the study shows there is still room to improve. That is where the next gains compound.

Try Krea 2

The Krea Team