Guides

What it takes to run ComfyUI and open-source AI models: the graphics card, the memory, the setup, and the options when your computer isn't big enough.

• ComfyUI GPU requirements: how much VRAM you needHow much graphics card memory (VRAM) ComfyUI needs for image, video and audio models, what fits on 8, 12, 16, 24 and 32 GB cards, system memory and disk space, and what to do when your card is too small.Read
• How to run ComfyUI without a good GPUYour options when your computer has no graphics card, a small one or a laptop GPU: low-VRAM settings, smaller compressed models, running on the processor, renting a cloud GPU by the hour, or by the second.Read
• Running ComfyUI locally: what it takesWhat installing and running ComfyUI on your own computer involves: the install options, drivers, model files and add-ons (custom nodes), disk space, and the errors people hit most, like CUDA out of memory and missing nodes.Read
• ComfyUI on a Mac: what works on Apple siliconRunning ComfyUI on an M1, M2, M3, M4 or M5 Mac: how unified memory helps, how fast it is next to an NVIDIA card, which models and files cause trouble, why video is slow, and what Intel Macs can do.Read
• Running ComfyUI on a cloud GPUHow to run ComfyUI and open models on a rented cloud GPU: hourly GPU computers (pods), per-second serverless GPUs, hosted ComfyUI websites, what each costs on RunPod, and where your files end up.Read