ComfyUI optimizes VRAM usage and expands cloud ecosystem

ComfyUI, a node-based interface for running generative image and video models, has undergone significant software updates in 2026 to address memory management issues. While users often face CUDA out-of-memory errors or slow processing due to limited video memory (VRAM), recent application changes have fixed several previous limitations.

There is no official minimum VRAM specification from Comfy Org; instead, requirements depend on specific workflows. Peak memory usage occurs when diffusion models, text encoders, and VAEs are resident simultaneously. For instance, a single image model at standard resolution typically requires 8 to 12 GB of VRAM, while workflows involving ControlNet, LoRA stacks, and upscaling can require between 12 and 24 GB.

In tandem with these technical developments, Comfy is expanding its ecosystem. The company is seeking a Senior Growth Engineer to lead product-led growth for Comfy Cloud, a zero-setup platform designed to facilitate community creation within the ComfyUI ecosystem.