The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Free memory before doing tiled decode.
comfyanonymous2865f913f754
diff --git a/comfy/sd.py b/comfy/sd.pyindex 7e76f6fa1..95fc6d271 100644--- a/comfy/sd.py+++ b/comfy/sd.py@@ -362,7 +362,8 @@ class VAE:return pixel_samplesdef decode_tiled(self, samples, tile_x=None, tile_y=None, overlap=None):- model_management.load_model_gpu(self.patcher)+ memory_used = self.memory_used_decode(samples.shape, self.vae_dtype) #TODO: calculate mem required for tile+ model_management.load_models_gpu([self.patcher], memory_required=memory_used)dims = samples.ndim - 2args = {}if tile_x is not None: