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.

Put clip vision outputs on the CPU.

comfyanonymous4e89b2c25a2e
diff --git a/comfy/clip_vision.py b/comfy/clip_vision.py
index 8635e577d..daaa2f2bf 100644
--- a/comfy/clip_vision.py
+++ b/comfy/clip_vision.py
@@ -50,6 +50,11 @@ class ClipVisionModel():
with precision_scope(comfy.model_management.get_autocast_device(self.load_device), torch.float32):
outputs = self.model(pixel_values=pixel_values)
+
+ for k in outputs:
+ t = outputs[k]
+ if t is not None:
+ outputs[k] = t.cpu()
return outputs
def convert_to_transformers(sd, prefix):