Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Add argument to run the VAE on the CPU.

comfyanonymous1b103e0cb2d7
diff --git a/comfy/cli_args.py b/comfy/cli_args.py
index 8de0adb53..50d7b62fa 100644
--- a/comfy/cli_args.py
+++ b/comfy/cli_args.py
@@ -66,6 +66,8 @@ fpvae_group.add_argument("--fp16-vae", action="store_true", help="Run the VAE in
fpvae_group.add_argument("--fp32-vae", action="store_true", help="Run the VAE in full precision fp32.")
fpvae_group.add_argument("--bf16-vae", action="store_true", help="Run the VAE in bf16.")
+parser.add_argument("--cpu-vae", action="store_true", help="Run the VAE on the CPU.")
+
fpte_group = parser.add_mutually_exclusive_group()
fpte_group.add_argument("--fp8_e4m3fn-text-enc", action="store_true", help="Store text encoder weights in fp8 (e4m3fn variant).")
fpte_group.add_argument("--fp8_e5m2-text-enc", action="store_true", help="Store text encoder weights in fp8 (e5m2 variant).")
diff --git a/comfy/model_management.py b/comfy/model_management.py
index c0cb4130c..fefd3c8c9 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -186,6 +186,9 @@ except:
if is_intel_xpu():
VAE_DTYPE = torch.bfloat16
+if args.cpu_vae:
+ VAE_DTYPE = torch.float32
+
if args.fp16_vae:
VAE_DTYPE = torch.float16
elif args.bf16_vae:
@@ -555,6 +558,8 @@ def intermediate_device():
return torch.device("cpu")
def vae_device():
+ if args.cpu_vae:
+ return torch.device("cpu")
return get_torch_device()
def vae_offload_device():