The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Don't add noise to the alpha channel in the Add Noise to Image node (#15626)
Christian Byrne7ba217d6b5ac
diff --git a/comfy_extras/nodes_images.py b/comfy_extras/nodes_images.pyindex be8988976..9949fd1bd 100644--- a/comfy_extras/nodes_images.py+++ b/comfy_extras/nodes_images.py@@ -336,6 +336,8 @@ class ImageAddNoise(IO.ComfyNode):def execute(cls, image, seed, strength) -> IO.NodeOutput:generator = torch.manual_seed(seed)s = torch.clip((image + strength * torch.randn(image.size(), generator=generator, device="cpu").to(image)), min=0.0, max=1.0)+ if image.shape[-1] == 4: # alpha stores transparency, not color+ s[..., 3] = image[..., 3]return IO.NodeOutput(s)repeat = execute # TODO: removediff --git a/tests-unit/comfy_extras_test/image_add_noise_alpha_test.py b/tests-unit/comfy_extras_test/image_add_noise_alpha_test.pynew file mode 100644index 000000000..3f71e834c--- /dev/null+++ b/tests-unit/comfy_extras_test/image_add_noise_alpha_test.py@@ -0,0 +1,63 @@+import torch++from comfy.cli_args import args as cli_args++if not torch.cuda.is_available():+ cli_args.cpu = True++from comfy_extras.nodes_images import ImageAddNoise # noqa: E402+++def image(channels, value=0.5, alpha=0.5, size=8):+ t = torch.full((1, size, size, channels), value)+ if channels == 4:+ t[..., 3] = alpha+ return t+++def test_rgb_gets_noise_on_every_channel():+ src = image(3)++ s = ImageAddNoise.execute(src, 0, 0.5).result[0]++ assert s.shape == src.shape+ assert not torch.equal(s, src)+++def test_rgba_keeps_alpha_untouched():+ src = image(4)++ s = ImageAddNoise.execute(src, 0, 0.5).result[0]++ assert s.shape[-1] == 4+ assert torch.equal(s[..., 3], src[..., 3])+ assert not torch.equal(s[..., :3], src[..., :3])+++def test_fully_transparent_pixels_stay_transparent():+ src = image(4, alpha=0.0)++ s = ImageAddNoise.execute(src, 0, 1.0).result[0]++ assert torch.all(s[..., 3] == 0.0)+++def test_only_the_alpha_channel_is_guarded():+ """Colour channels must still get exactly the unguarded noise result."""+ src = image(4)+ generator = torch.manual_seed(0)+ unguarded = torch.clip(src + 0.5 * torch.randn(src.size(), generator=generator), min=0.0, max=1.0)++ s = ImageAddNoise.execute(src, 0, 0.5).result[0]++ assert torch.equal(s[..., :3], unguarded[..., :3])+ assert torch.equal(s[..., 3], src[..., 3])+++def test_does_not_mutate_input():+ src = image(4)+ before = src.clone()++ ImageAddNoise.execute(src, 0, 0.5)++ assert torch.equal(src, before)