298977887/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-04-20 03:13:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'tiled-progress' of https://github.com/pythongosssss/ComfyUI

Browse Source
This commit is contained in:
comfyanonymous 2023-05-03 16:24:56 -04:00
commit a74e176a24
3 changed files with 28 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
comfy/sd.py
View File

@ -514,11 +514,14 @@ class VAE:
self.device = device self.device = device
def decode_tiled_(self, samples, tile_x=64, tile_y=64, overlap = 16): def decode_tiled_(self, samples, tile_x=64, tile_y=64, overlap = 16):
steps = samples.shape[0] * utils.get_tiled_scale_steps(samples.shape[3], samples.shape[2], tile_x, tile_y, overlap)
pbar = utils.ProgressBar(steps)
decode_fn = lambda a: (self.first_stage_model.decode(1. / self.scale_factor * a.to(self.device)) + 1.0) decode_fn = lambda a: (self.first_stage_model.decode(1. / self.scale_factor * a.to(self.device)) + 1.0)
output = torch.clamp(( output = torch.clamp((
(utils.tiled_scale(samples, decode_fn, tile_x // 2, tile_y * 2, overlap, upscale_amount = 8) + (utils.tiled_scale(samples, decode_fn, tile_x // 2, tile_y * 2, overlap, upscale_amount = 8, pbar = pbar) +
utils.tiled_scale(samples, decode_fn, tile_x * 2, tile_y // 2, overlap, upscale_amount = 8) + utils.tiled_scale(samples, decode_fn, tile_x * 2, tile_y // 2, overlap, upscale_amount = 8, pbar = pbar) +
utils.tiled_scale(samples, decode_fn, tile_x, tile_y, overlap, upscale_amount = 8)) utils.tiled_scale(samples, decode_fn, tile_x, tile_y, overlap, upscale_amount = 8, pbar = pbar))
/ 3.0) / 2.0, min=0.0, max=1.0) / 3.0) / 2.0, min=0.0, max=1.0)
return output return output
@ -562,9 +565,13 @@ class VAE:
model_management.unload_model() model_management.unload_model()
self.first_stage_model = self.first_stage_model.to(self.device) self.first_stage_model = self.first_stage_model.to(self.device)
pixel_samples = pixel_samples.movedim(-1,1).to(self.device) pixel_samples = pixel_samples.movedim(-1,1).to(self.device)
samples = utils.tiled_scale(pixel_samples, lambda a: self.first_stage_model.encode(2. * a - 1.).sample() * self.scale_factor, tile_x, tile_y, overlap, upscale_amount = (1/8), out_channels=4)
samples += utils.tiled_scale(pixel_samples, lambda a: self.first_stage_model.encode(2. * a - 1.).sample() * self.scale_factor, tile_x * 2, tile_y // 2, overlap, upscale_amount = (1/8), out_channels=4) steps = utils.get_tiled_scale_steps(pixel_samples.shape[3], pixel_samples.shape[2], tile_x, tile_y, overlap)
samples += utils.tiled_scale(pixel_samples, lambda a: self.first_stage_model.encode(2. * a - 1.).sample() * self.scale_factor, tile_x // 2, tile_y * 2, overlap, upscale_amount = (1/8), out_channels=4) pbar = utils.ProgressBar(steps)
samples = utils.tiled_scale(pixel_samples, lambda a: self.first_stage_model.encode(2. * a - 1.).sample() * self.scale_factor, tile_x, tile_y, overlap, upscale_amount = (1/8), out_channels=4, pbar=pbar)
samples += utils.tiled_scale(pixel_samples, lambda a: self.first_stage_model.encode(2. * a - 1.).sample() * self.scale_factor, tile_x * 2, tile_y // 2, overlap, upscale_amount = (1/8), out_channels=4, pbar=pbar)
samples += utils.tiled_scale(pixel_samples, lambda a: self.first_stage_model.encode(2. * a - 1.).sample() * self.scale_factor, tile_x // 2, tile_y * 2, overlap, upscale_amount = (1/8), out_channels=4, pbar=pbar)
samples /= 3.0 samples /= 3.0
self.first_stage_model = self.first_stage_model.cpu() self.first_stage_model = self.first_stage_model.cpu()
samples = samples.cpu() samples = samples.cpu()

10
comfy/utils.py
View File

@ -62,8 +62,14 @@ def common_upscale(samples, width, height, upscale_method, crop):
s = samples s = samples
return torch.nn.functional.interpolate(s, size=(height, width), mode=upscale_method) return torch.nn.functional.interpolate(s, size=(height, width), mode=upscale_method)
def get_tiled_scale_steps(width, height, tile_x, tile_y, overlap):
it_1 = -(height // -(tile_y * 2 - overlap)) * -(width // -(tile_x // 2 - overlap))
it_2 = -(height // -(tile_y // 2 - overlap)) * -(width // -(tile_x * 2 - overlap))
it_3 = -(height // -(tile_y - overlap)) * -(width // -(tile_x - overlap))
return it_1 + it_2 + it_3
@torch.inference_mode() @torch.inference_mode()
def tiled_scale(samples, function, tile_x=64, tile_y=64, overlap = 8, upscale_amount = 4, out_channels = 3): def tiled_scale(samples, function, tile_x=64, tile_y=64, overlap = 8, upscale_amount = 4, out_channels = 3, pbar = None):
output = torch.empty((samples.shape[0], out_channels, round(samples.shape[2] * upscale_amount), round(samples.shape[3] * upscale_amount)), device="cpu") output = torch.empty((samples.shape[0], out_channels, round(samples.shape[2] * upscale_amount), round(samples.shape[3] * upscale_amount)), device="cpu")
for b in range(samples.shape[0]): for b in range(samples.shape[0]):
s = samples[b:b+1] s = samples[b:b+1]
@ -83,6 +89,8 @@ def tiled_scale(samples, function, tile_x=64, tile_y=64, overlap = 8, upscale_am
mask[:,:,:,mask.shape[3]- 1 - t: mask.shape[3]- t] *= ((1.0/feather) * (t + 1)) mask[:,:,:,mask.shape[3]- 1 - t: mask.shape[3]- t] *= ((1.0/feather) * (t + 1))
out[:,:,round(y*upscale_amount):round((y+tile_y)*upscale_amount),round(x*upscale_amount):round((x+tile_x)*upscale_amount)] += ps * mask out[:,:,round(y*upscale_amount):round((y+tile_y)*upscale_amount),round(x*upscale_amount):round((x+tile_x)*upscale_amount)] += ps * mask
out_div[:,:,round(y*upscale_amount):round((y+tile_y)*upscale_amount),round(x*upscale_amount):round((x+tile_x)*upscale_amount)] += mask out_div[:,:,round(y*upscale_amount):round((y+tile_y)*upscale_amount),round(x*upscale_amount):round((x+tile_x)*upscale_amount)] += mask
if pbar is not None:
pbar.update(1)
output[b:b+1] = out/out_div output[b:b+1] = out/out_div
return output return output

7
comfy_extras/nodes_upscale_model.py
View File

@ -37,7 +37,12 @@ class ImageUpscaleWithModel:
device = model_management.get_torch_device() device = model_management.get_torch_device()
upscale_model.to(device) upscale_model.to(device)
in_img = image.movedim(-1,-3).to(device) in_img = image.movedim(-1,-3).to(device)
s = comfy.utils.tiled_scale(in_img, lambda a: upscale_model(a), tile_x=128 + 64, tile_y=128 + 64, overlap = 8, upscale_amount=upscale_model.scale)
tile = 128 + 64
overlap = 8
steps = -(in_img.shape[2] // -(tile - overlap)) * -(in_img.shape[3] // -(tile - overlap))
pbar = comfy.utils.ProgressBar(steps)
s = comfy.utils.tiled_scale(in_img, lambda a: upscale_model(a), tile_x=tile, tile_y=tile, overlap=overlap, upscale_amount=upscale_model.scale, pbar=pbar)
upscale_model.cpu() upscale_model.cpu()
s = torch.clamp(s.movedim(-3,-1), min=0, max=1.0) s = torch.clamp(s.movedim(-3,-1), min=0, max=1.0)
return (s,) return (s,)