VRAM
-
ComfyUI GPU Requirements: How Much VRAM You Need
Published weight sizes for SD 1.5, SDXL and Flux mapped to real VRAM tiers, plus the system RAM, disk and runtime flags that lower the floor.
-
ComfyUI Not Enough GPU Memory: Fix CUDA Out of Memory
Learn which stage ran out of VRAM, from model load to VAE decode, and the fixes for each: batch and resolution, fp8 weights, --cpu-vae and --cache-none.
-
ComfyUI vs Automatic1111 vs Forge: Which to Run
Release cadence, memory architecture, extension risk and automation compared across ComfyUI, the AUTOMATIC1111 WebUI and Forge, with a decision table.
-
How ComfyUI Graphs Execute and Where VRAM Goes
Node graph execution order, result caching, VRAM pressure across the three failure stages, and custom node dependency management in ComfyUI.