xinyun/ComfyUI-KJNodes
1
0
Fork 0
mirror of https://git.datalinker.icu/kijai/ComfyUI-KJNodes.git synced 2025-12-22 11:14:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

add batched encode/decode

Browse Source
This commit is contained in:
kijai 2024-02-25 19:04:28 +02:00
parent 0cb0f87a6b
commit 6829b0ccf8
1 changed files with 17 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
nodes.py
View File

@ -3686,6 +3686,7 @@ class Intrinsic_lora_sampling:
"clip": ("CLIP", ), "clip": ("CLIP", ),
"vae": ("VAE", ), "vae": ("VAE", ),
"image": ("IMAGE",), "image": ("IMAGE",),
"per_batch": ("INT", {"default": 16, "min": 1, "max": 4096, "step": 1}),
}} }}
@ -3693,14 +3694,17 @@ class Intrinsic_lora_sampling:
FUNCTION = "onestepsample" FUNCTION = "onestepsample"
CATEGORY = "KJNodes" CATEGORY = "KJNodes"
def onestepsample(self, model, lora_name, clip, vae, image, text, task): def onestepsample(self, model, lora_name, clip, vae, image, text, task, per_batch):
pbar = comfy.utils.ProgressBar(3) pbar = comfy.utils.ProgressBar(3)
encoded_latent, = VAEEncode.encode(self, vae, image[:,:,:,:3])
sample = encoded_latent["samples"]
noise = torch.zeros(sample.size(), dtype=sample.dtype, layout=sample.layout, device="cpu")
image_list = []
for start_idx in range(0, image.shape[0], per_batch):
sub_pixels = vae.vae_encode_crop_pixels(image[start_idx:start_idx+per_batch])
image_list.append(vae.encode(sub_pixels[:,:,:,:3]))
sample = torch.cat(image_list, dim=0)
noise = torch.zeros(sample.size(), dtype=sample.dtype, layout=sample.layout, device="cpu")
prompt = task + "," + text prompt = task + "," + text
print(prompt)
positive, = CLIPTextEncode.encode(self, clip, prompt) positive, = CLIPTextEncode.encode(self, clip, prompt)
pbar.update(1) pbar.update(1)
negative = positive #negative shouldn't do anything in this scenario negative = positive #negative shouldn't do anything in this scenario
@ -3726,8 +3730,15 @@ class Intrinsic_lora_sampling:
denoise=1.0, disable_noise=True, start_step=0, last_step=1, denoise=1.0, disable_noise=True, start_step=0, last_step=1,
force_full_denoise=True, noise_mask=None, callback=None, disable_pbar=True, seed=None)} force_full_denoise=True, noise_mask=None, callback=None, disable_pbar=True, seed=None)}
pbar.update(1) pbar.update(1)
image_out, = VAEDecode.decode(self, vae, samples)
decoded = []
for start_idx in range(0, samples["samples"].shape[0], per_batch):
decoded.append(vae.decode(samples["samples"][start_idx:start_idx+per_batch]))
image_out = torch.cat(decoded, dim=0)
#image_out, = VAEDecode.decode(self, vae, samples)
pbar.update(1) pbar.update(1)
if task == 'depth map': if task == 'depth map':
imax = image_out.max() imax = image_out.max()
imin = image_out.min() imin = image_out.min()