The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Use the same units for tile size in VAEDecodeTiled and VAEEncodeTiled.
comfyanonymous8ddd081b093f
diff --git a/nodes.py b/nodes.pyindex 81eeda042..3e4d5240b 100644--- a/nodes.py+++ b/nodes.py@@ -245,7 +245,7 @@ class VAEDecodeTiled:@classmethoddef INPUT_TYPES(s):return {"required": {"samples": ("LATENT", ), "vae": ("VAE", ),- "tile_size": ("INT", {"default": 64, "min": 64, "max": 4096, "step": 64})+ "tile_size": ("INT", {"default": 512, "min": 192, "max": 4096, "step": 64})}}RETURN_TYPES = ("IMAGE",)FUNCTION = "decode"@@ -253,7 +253,7 @@ class VAEDecodeTiled:CATEGORY = "_for_testing"def decode(self, vae, samples, tile_size):- return (vae.decode_tiled(samples["samples"], tile_x=tile_size, tile_y=tile_size, ), )+ return (vae.decode_tiled(samples["samples"], tile_x=tile_size // 8, tile_y=tile_size // 8, ), )class VAEEncode:@classmethod@@ -283,7 +283,7 @@ class VAEEncodeTiled:@classmethoddef INPUT_TYPES(s):return {"required": {"pixels": ("IMAGE", ), "vae": ("VAE", ),- "tile_size": ("INT", {"default": 512, "min": 192, "max": 4096, "step": 64})+ "tile_size": ("INT", {"default": 512, "min": 320, "max": 4096, "step": 64})}}RETURN_TYPES = ("LATENT",)FUNCTION = "encode"