The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
make control-net load order deterministic (#13701)
rattus1ac78180b3f7
diff --git a/comfy/model_management.py b/comfy/model_management.pyindex 02ad66656..21738a4c7 100644--- a/comfy/model_management.py+++ b/comfy/model_management.py@@ -721,13 +721,15 @@ def load_models_gpu(models, memory_required=0, force_patch_weights=False, minimuelse:minimum_memory_required = max(inference_memory, minimum_memory_required + extra_reserved_memory())- models_temp = set()+ # Order-preserving dedup. A plain set() would randomize iteration order across runs+ models_temp = {}for m in models:- models_temp.add(m)+ models_temp[m] = Nonefor mm in m.model_patches_models():- models_temp.add(mm)+ models_temp[mm] = None- models = models_temp+ models = list(models_temp)+ models.reverse()models_to_load = []diff --git a/comfy/sampler_helpers.py b/comfy/sampler_helpers.pyindex bbba09e26..3782fd2d5 100644--- a/comfy/sampler_helpers.py+++ b/comfy/sampler_helpers.py@@ -89,7 +89,8 @@ def get_additional_models(conds, dtype):gligen += get_models_from_cond(conds[k], "gligen")add_models += get_models_from_cond(conds[k], "additional_models")- control_nets = set(cnets)+ # Order-preserving dedup. A plain set() would randomize iteration order across runs+ control_nets = list(dict.fromkeys(cnets))inference_memory = 0control_models = []