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.

Fix potential lowvram issue.

comfyanonymous83f343146ae1
diff --git a/comfy/model_patcher.py b/comfy/model_patcher.py
index 1edbf24ab..54a64aa47 100644
--- a/comfy/model_patcher.py
+++ b/comfy/model_patcher.py
@@ -367,7 +367,7 @@ class ModelPatcher:
if mem_counter + module_mem >= lowvram_model_memory:
lowvram_weight = True
lowvram_counter += 1
- if m.comfy_cast_weights:
+ if hasattr(m, "prev_comfy_cast_weights"): #Already lowvramed
continue
weight_key = "{}.weight".format(n)