Local inference benchmark

Turbo image models on a single NVIDIA GeForce RTX 3090

Models Z-Image Turbo, Boogu Turbo, Krea 2 Turbo Resolution 1024²Date 2026-06-23

Three open-weight Turbo image models benchmarked head-to-head on one RTX 3090 — identical prompt, seed, and resolution, with per-step throughput, real-world seconds-per-image, VRAM footprint, and power measured automatically over 10 runs each. Fastest per image: Boogu Turbo at 5.54 s.

Interactive

Performance dashboard

Throughput, seconds-per-image, VRAM headroom, power and run-to-run consistency as interactive charts.

Open dashboard →

Visual

Image comparison report

The same prompt and seed across all three models, plus a variety grid of every generated image per model.

Open report →