TL;DR: A community test shows the MOSS image diffusion model producing cinematic 4MP renders in under a minute on consumer hardware, with a stock ComfyUI setup.
Summary: A Reddit user benchmarked MOSS image (moss_image_sft_int8_convrot.safetensors) using a plain ComfyUI template, paired with a Qwen3.5-9B int8 text encoder and the Flux2 VAE at 50 steps, CFG 3, and euler/sgm_uniform sampling. At those settings, 4MP generations completed in roughly 55 seconds with 19.5GB peak VRAM on a 5090-class GPU, and the author noted a distinctly cinematic look straight out of the box. The poster corrected the original title: the resolution is 4MP, not 4K.
Why it matters: It gives builders a concrete reference point for MOSS image's quality-per-second and VRAM footprint on high-end consumer cards. Worth replicating the config on your own GPU before committing it to a production pipeline.
Source: reddit