Fix all comparisons against the DEFAULT_PRECISION constant. DEFAULT_PRECISION is a torch.dtype. Previously, it was compared to a str in a number of places where it would always resolve to False. This is a bugfix that results in a change to the default behavior. In practice, this will not change the behavior for many users, because it only causes a change in behavior if a users has configured float32 as their default precision.

Author: RyanJDick

invokeai/app/invocations/constants.py

@@ -1,6 +1,7 @@
 from typing import Literal
 
 from invokeai.backend.stable_diffusion.schedulers import SCHEDULER_MAP
+from invokeai.backend.util.devices import TorchDevice
 
 LATENT_SCALE_FACTOR = 8
 """
@@ -15,3 +16,5 @@
 
 IMAGE_MODES = Literal["L", "RGB", "RGBA", "CMYK", "YCbCr", "LAB", "HSV", "I", "F"]
 """A literal type for PIL image modes supported by Invoke"""
+
+DEFAULT_PRECISION = TorchDevice.choose_torch_dtype()

invokeai/app/invocations/create_denoise_mask.py

@@ -6,7 +6,7 @@
 from torchvision.transforms.functional import resize as tv_resize
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
-from invokeai.app.invocations.denoise_latents import DEFAULT_PRECISION
+from invokeai.app.invocations.constants import DEFAULT_PRECISION
 from invokeai.app.invocations.fields import FieldDescriptions, ImageField, Input, InputField
 from invokeai.app.invocations.image_to_latents import ImageToLatentsInvocation
 from invokeai.app.invocations.model import VAEField
@@ -30,7 +30,7 @@ class CreateDenoiseMaskInvocation(BaseInvocation):
     mask: ImageField = InputField(description="The mask to use when pasting", ui_order=2)
     tiled: bool = InputField(default=False, description=FieldDescriptions.tiled, ui_order=3)
     fp32: bool = InputField(
-        default=DEFAULT_PRECISION == "float32",
+        default=DEFAULT_PRECISION == torch.float32,
         description=FieldDescriptions.fp32,
         ui_order=4,
     )

invokeai/app/invocations/create_gradient_mask.py

@@ -7,7 +7,7 @@
 from torchvision.transforms.functional import resize as tv_resize
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, BaseInvocationOutput, invocation, invocation_output
-from invokeai.app.invocations.denoise_latents import DEFAULT_PRECISION
+from invokeai.app.invocations.constants import DEFAULT_PRECISION
 from invokeai.app.invocations.fields import (
     DenoiseMaskField,
     FieldDescriptions,
@@ -74,7 +74,7 @@ class CreateGradientMaskInvocation(BaseInvocation):
     )
     tiled: bool = InputField(default=False, description=FieldDescriptions.tiled, ui_order=8)
     fp32: bool = InputField(
-        default=DEFAULT_PRECISION == "float32",
+        default=DEFAULT_PRECISION == torch.float32,
         description=FieldDescriptions.fp32,
         ui_order=9,
     )

invokeai/app/invocations/denoise_latents.py

@@ -59,8 +59,6 @@
 from .controlnet_image_processors import ControlField
 from .model import ModelIdentifierField, UNetField
 
-DEFAULT_PRECISION = TorchDevice.choose_torch_dtype()
-
 
 def get_scheduler(
     context: InvocationContext,

invokeai/app/invocations/image_to_latents.py

@@ -12,7 +12,7 @@
 from diffusers.models.autoencoders.autoencoder_tiny import AutoencoderTiny
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
-from invokeai.app.invocations.denoise_latents import DEFAULT_PRECISION
+from invokeai.app.invocations.constants import DEFAULT_PRECISION
 from invokeai.app.invocations.fields import (
     FieldDescriptions,
     ImageField,
@@ -44,7 +44,7 @@ class ImageToLatentsInvocation(BaseInvocation):
         input=Input.Connection,
     )
     tiled: bool = InputField(default=False, description=FieldDescriptions.tiled)
-    fp32: bool = InputField(default=DEFAULT_PRECISION == "float32", description=FieldDescriptions.fp32)
+    fp32: bool = InputField(default=DEFAULT_PRECISION == torch.float32, description=FieldDescriptions.fp32)
 
     @staticmethod
     def vae_encode(vae_info: LoadedModel, upcast: bool, tiled: bool, image_tensor: torch.Tensor) -> torch.Tensor:

invokeai/app/invocations/latents_to_image.py

@@ -11,7 +11,7 @@
 from diffusers.models.unets.unet_2d_condition import UNet2DConditionModel
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
-from invokeai.app.invocations.denoise_latents import DEFAULT_PRECISION
+from invokeai.app.invocations.constants import DEFAULT_PRECISION
 from invokeai.app.invocations.fields import (
     FieldDescriptions,
     Input,
@@ -46,7 +46,7 @@ class LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
         input=Input.Connection,
     )
     tiled: bool = InputField(default=False, description=FieldDescriptions.tiled)
-    fp32: bool = InputField(default=DEFAULT_PRECISION == "float32", description=FieldDescriptions.fp32)
+    fp32: bool = InputField(default=DEFAULT_PRECISION == torch.float32, description=FieldDescriptions.fp32)
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> ImageOutput: