Quantized Krea 2 Boosts ComfyUI Inference Speed by Over 2x

A community-driven quantization project for Krea 2 utilizes SVDQuant and INT8/W4A4 formats to accelerate ComfyUI inference by up to 2.1 times. This approach significantly speeds up local workflows while maintaining nearly identical generation quality.

2026-07-25 ~ 2026-07-27 · 2 related posts

• Krea 2 SVDQuant checkpoints cut ComfyUI runtime about 2× on NVIDIA GPUs — LightAppropriate624 · 2026-07-25
• SVDQuant and INT8/W4A4 make Krea 2 up to 2.1× faster in ComfyUI — LightAppropriate624 · 2026-07-27