192 evaluations across 6 categories
Benchmark V1 verdicts are produced by VLM judges and can contain mistakes. Treat PASS/FAIL labels as machine-assisted assessments, and inspect the images yourself. Learn more about the methodology.
Source-backed model context, size, cost, and request settings for this ImageBench V1 run.
Qwen-Image-2512 is the December update of Qwen's text-to-image foundation model, improving human realism, natural detail, and text rendering over the earlier Qwen-Image release.
Reve 2.1 is Reve AI's flagship text-to-image model, generating natively at 4K (4096x4096) with layout intelligence — it reasons about structure and spatial hierarchy before generating. It is a standout at in-image typography (including foreign scripts) and ranks #2 on the Text-to-Image Arena, well suited to posters, packaging, and marketing design. The older fal-ai/reve/text-to-image endpoint serves the earlier Reve Image 1.0.
On ImageBench V1 (192 prompts, 6 categories), Qwen Image 2512 20b scores 59.6 overall with a 52.1% task pass rate, while Reve 2.1 scores 74.8 overall with a 79.7% pass rate. The biggest difference is graphic design: Qwen Image 2512 20b passes 25% of those prompts vs 87.5% for Reve 2.1. Qwen Image 2512 20b is an open-weight model you can run locally; Reve 2.1 is an API model ($0.04 / image).
Reve 2.1 is better overall on ImageBench V1, scoring 74.8 vs 59.6. Qwen Image 2512 20b is strongest at professional studio photography (81.5%). Reve 2.1 is strongest at text rendering (100%).
Yes — on the Graphical design category of ImageBench V1, Qwen Image 2512 20b passes 25% of prompts and Reve 2.1 passes 87.5%. That is the largest per-category difference between Qwen Image 2512 20b and Reve 2.1.
The ImageBench gallery shows Qwen Image 2512 20b and Reve 2.1 outputs for the same 192 prompts side by side, filterable by category and difficulty — the fastest way to judge them with your own eyes instead of trusting aggregate scores. Open them in the gallery.