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.

Lower minimum ratio of loaded weights on Nvidia.

comfyanonymous255edf22463f
diff --git a/comfy/model_management.py b/comfy/model_management.py
index cd7389c42..225a83e05 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -218,7 +218,7 @@ def is_amd():
MIN_WEIGHT_MEMORY_RATIO = 0.4
if is_nvidia():
- MIN_WEIGHT_MEMORY_RATIO = 0.2
+ MIN_WEIGHT_MEMORY_RATIO = 0.1
ENABLE_PYTORCH_ATTENTION = False
if args.use_pytorch_cross_attention: