fix dimension errors when inpainting model is used with hires-fix (#2440)

This commit is contained in:
Kevin Turner 2023-01-31 11:04:02 -08:00 committed by GitHub
commit bde5874707
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -3,10 +3,10 @@ ldm.invoke.generator.txt2img inherits from ldm.invoke.generator
'''
import math
from diffusers.utils.logging import get_verbosity, set_verbosity, set_verbosity_error
from typing import Callable, Optional
import torch
from diffusers.utils.logging import get_verbosity, set_verbosity, set_verbosity_error
from ldm.invoke.generator.base import Generator
from ldm.invoke.generator.diffusers_pipeline import trim_to_multiple_of, StableDiffusionGeneratorPipeline, \
@ -128,18 +128,13 @@ class Txt2Img2Img(Generator):
scaled_width = width
scaled_height = height
device = self.model.device
device = self.model.device
channels = self.latent_channels
if channels == 9:
channels = 4 # we don't really want noise for all the mask channels
shape = (1, channels,
scaled_height // self.downsampling_factor, scaled_width // self.downsampling_factor)
if self.use_mps_noise or device.type == 'mps':
return torch.randn([1,
self.latent_channels,
scaled_height // self.downsampling_factor,
scaled_width // self.downsampling_factor],
dtype=self.torch_dtype(),
device='cpu').to(device)
return torch.randn(shape, dtype=self.torch_dtype(), device='cpu').to(device)
else:
return torch.randn([1,
self.latent_channels,
scaled_height // self.downsampling_factor,
scaled_width // self.downsampling_factor],
dtype=self.torch_dtype(),
device=device)
return torch.randn(shape, dtype=self.torch_dtype(), device=device)