andreas/ComfyUI
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

WIP way to support multi multi dimensional latents. (#10456)

Browse Source
This commit is contained in:
comfyanonymous
2025-10-23 18:21:14 -07:00
committed by GitHub
parent a1864c01f2
commit 1bcda6df98
5 changed files with 158 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
comfy/sample.py
View File

@@ -4,13 +4,9 @@ import comfy.samplers
import comfy.utils
import numpy as np
import logging
import comfy.nested_tensor
def prepare_noise(latent_image, seed, noise_inds=None):
"""
creates random noise given a latent image and a seed.
optional arg skip can be used to skip and discard x number of noise generations for a given seed
"""
generator = torch.manual_seed(seed)
def prepare_noise_inner(latent_image, generator, noise_inds=None):
if noise_inds is None:
return torch.randn(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu")
@@ -22,9 +18,28 @@ def prepare_noise(latent_image, seed, noise_inds=None):
noises.append(noise)
noises = [noises[i] for i in inverse]
noises = torch.cat(noises, axis=0)
def prepare_noise(latent_image, seed, noise_inds=None):
"""
creates random noise given a latent image and a seed.
optional arg skip can be used to skip and discard x number of noise generations for a given seed
"""
generator = torch.manual_seed(seed)
if latent_image.is_nested:
tensors = latent_image.unbind()
noises = []
for t in tensors:
noises.append(prepare_noise_inner(t, generator, noise_inds))
noises = comfy.nested_tensor.NestedTensor(noises)
else:
noises = prepare_noise_inner(latent_image, generator, noise_inds)
return noises
def fix_empty_latent_channels(model, latent_image):
if latent_image.is_nested:
return latent_image
latent_format = model.get_model_object("latent_format") #Resize the empty latent image so it has the right number of channels
if latent_format.latent_channels != latent_image.shape[1] and torch.count_nonzero(latent_image) == 0:
latent_image = comfy.utils.repeat_to_batch_size(latent_image, latent_format.latent_channels, dim=1)