Mirrored_Repos/InvokeAI
1
0
Fork 0
mirror of https://github.com/invoke-ai/InvokeAI synced 2024-08-30 20:32:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

fixing aspect ratio on hires

Browse Source
This commit is contained in:
tyler 2022-10-25 20:59:13 -05:00
parent 4104ac6270
commit 62412f8398
1 changed files with 13 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
ldm/invoke/generator/txt2img2img.py
View File

@ -25,28 +25,27 @@ class Txt2Img2Img(Generator):
uc, c = conditioning uc, c = conditioning
@torch.no_grad() @torch.no_grad()
def make_image(x_T): def make_image(x_T):
trained_square = 512 * 512
actual_square = width * height
scale = math.sqrt(trained_square / actual_square)
scale_dim = min(width, height)
scale = 512 / scale_dim
init_width = math.ceil(scale * width / 64) * 64 init_width = math.ceil(scale * width / 64) * 64
init_height = math.ceil(scale * height / 64) * 64 init_height = math.ceil(scale * height / 64) * 64
shape = [ shape = [
self.latent_channels, self.latent_channels,
init_height // self.downsampling_factor, init_height // self.downsampling_factor,
init_width // self.downsampling_factor, init_width // self.downsampling_factor,
] ]
sampler.make_schedule( sampler.make_schedule(
ddim_num_steps=steps, ddim_eta=ddim_eta, verbose=False ddim_num_steps=steps, ddim_eta=ddim_eta, verbose=False
) )
#x = self.get_noise(init_width, init_height) #x = self.get_noise(init_width, init_height)
x = x_T x = x_T
if self.free_gpu_mem and self.model.model.device != self.model.device: if self.free_gpu_mem and self.model.model.device != self.model.device:
self.model.model.to(self.model.device) self.model.model.to(self.model.device)
@ -62,15 +61,15 @@ class Txt2Img2Img(Generator):
eta = ddim_eta, eta = ddim_eta,
img_callback = step_callback img_callback = step_callback
) )
print( print(
f"\n>> Interpolating from {init_width}x{init_height} to {width}x{height} using DDIM sampling" f"\n>> Interpolating from {init_width}x{init_height} to {width}x{height} using DDIM sampling"
) )
# resizing # resizing
samples = torch.nn.functional.interpolate( samples = torch.nn.functional.interpolate(
samples, samples,
size=(height // self.downsampling_factor, width // self.downsampling_factor), size=(height // self.downsampling_factor, width // self.downsampling_factor),
mode="bilinear" mode="bilinear"
) )
@ -116,7 +115,7 @@ class Txt2Img2Img(Generator):
else: else:
scaled_width = width scaled_width = width
scaled_height = height scaled_height = height
device = self.model.device device = self.model.device
if self.use_mps_noise or device.type == 'mps': if self.use_mps_noise or device.type == 'mps':
return torch.randn([1, return torch.randn([1,