Image model speed benchmark on RTX 4090: FLUX 3 Klein fastest at 4.3s/image

Long-Conversation-50 · reddit · 2026-11-21

A Reddit user benchmarked 11 image generation models on an RTX 4090 24GB with ComfyUI 0.45.0, using 7 prompts × 2 seeds per model: FLUX 3 Klein (int8, 15 steps) is fastest at 4.3s per 1008×1008 image; Z-Image Turbo 5.0s, Krea 2 Turbo 5.9s, Qwen Image 2.1 7.5s, MOSS Image 10.9s, MiniMax H3 Image 11.0s at 2016×2016 with only 6 steps. Slowest: HiDream O1 at 91.4s (40 steps + Gemma prompt writer + Z-Image refine) and Ideogram 4 at 30.6s, which also hit safety-filter blocks on 6 of 12 images. Edit-mode test limited to Qwen Image 2.1 Edit, MiniMax H3 Image, FIBO 2 Edit and Ming Image.

Original post →

More from Multimodal

Multimodal channel →