To run ComfyUI in low VRAM mode, start it with --lowvram; if that is not enough, use --novram.
python main.py --lowvram
What the flags do
These are the descriptions from ComfyUI's own cli_args.py:
| Flag | Help text |
|---|---|
--lowvram | Does nothing if dynamic VRAM is enabled. If it is not, makes the text encoders run on the CPU |
--novram | "When lowvram isn't enough" |
--highvram | Keeps models in GPU memory instead of unloading them to CPU memory after use |
--gpu-only | Stores and runs everything, including text encoders, on the GPU |
--cpu | Uses the CPU for everything (slow) |
The five are mutually exclusive: pass only one. The README says ComfyUI can run the biggest open models on as low as 4GB of VRAM plus 8GB of RAM with its weight streaming, so many "low VRAM" setups work without any flag. Check dynamic VRAM first, because it changes what --lowvram does.
Still failing? See the out of memory fixes and how much VRAM ComfyUI needs. The ComfyUI custom node install guide covers what to install once it runs.
Sources
- ComfyUI cli_args.py (VRAM flag group). github.com
- ComfyUI README. github.com