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.

Text encoder should initially load on the offload_device not the regular.

comfyanonymous1300a1bb4c87
diff --git a/comfy/sd.py b/comfy/sd.py
index 1edc5fd29..a63a0d1de 100644
--- a/comfy/sd.py
+++ b/comfy/sd.py
@@ -80,7 +80,7 @@ class CLIP:
load_device = model_management.text_encoder_device()
offload_device = model_management.text_encoder_offload_device()
- params['device'] = load_device
+ params['device'] = offload_device
if model_management.should_use_fp16(load_device, prioritize_performance=False):
params['dtype'] = torch.float16
else: