LoRA and no custom diffusers
This commit is contained in:
nullquant
+1
-1
@@ -1,4 +1,4 @@
|
||||
__pycache__/
|
||||
**/__pycache__/
|
||||
.vscode/
|
||||
*.tmp
|
||||
*.dblite
|
||||
|
||||
@@ -15,24 +15,32 @@ pip install -r requirements.txt
|
||||
```
|
||||
|
||||
Checkpoints of BrushNet can be downloaded from [here](https://drive.google.com/drive/folders/1fqmS1CEOvXCxNWFrsSYd_jHYXxrydh1n?usp=drive_link).
|
||||
Both `segmentation_mask_brushnet_ckpt` and `random_mask_brushnet_ckpt` should be placed inside `models/inpaint/brushnet` folder.
|
||||
Base model should be SD1.5 type (realisticVisionV60B1_v51VAE, for example).
|
||||
|
||||
The checkpoint in `segmentation_mask_brushnet_ckpt` provides checkpoints trained on BrushData, which has segmentation prior (mask are with the same shape of objects). The `random_mask_brushnet_ckpt` provides a more general ckpt for random mask shape.
|
||||
|
||||
|
||||
Both `segmentation_mask_brushnet_ckpt` and `random_mask_brushnet_ckpt` or only one of them should be placed inside `models/inpaint/brushnet` folder.
|
||||
|
||||
Base model should be SD1.5 type (realisticVisionV60B1_v51VAE, for example).
|
||||
|
||||
This is complete rework of previous version. There is no more custom `diffusers` library.
|
||||
There is no loadings from github. I also changed the workflow so now it is possible to use LoRa models.
|
||||
To users who already have previoous version: please, uninstall `diffusers` library before installation:
|
||||
|
||||
```
|
||||
pip uninstall diffusers
|
||||
```
|
||||
|
||||
Then install as usual.
|
||||
|
||||
## Usage
|
||||
|
||||
Below is an example for the intended workflow. The [json file](example/BrushNet_example.json) for the example can be found inside the 'example' directory.
|
||||
Below is an example for the intended workflow. The [json file](example/BrushNet_basic.json) for the example can be found inside the 'example' directory.
|
||||
|
||||
|
||||
|
||||
|
||||
This is example with LoRA: [json file](example/BrushNet_with_LoRA.json).
|
||||
|
||||
## Bugs
|
||||
|
||||
1. Nodes use custom `diffusers` library. It can potentially conflicts with other custom nodes.
|
||||
2. Sometimes `BrushNetPipeline` freezes on loading. Looking into the problem.
|
||||
|
||||
|
||||
|
||||
## Credits
|
||||
|
||||
@@ -0,0 +1,58 @@
|
||||
{
|
||||
"_class_name": "BrushNetModel",
|
||||
"_diffusers_version": "0.27.0.dev0",
|
||||
"_name_or_path": "runs/logs/brushnet_randommask/checkpoint-100000",
|
||||
"act_fn": "silu",
|
||||
"addition_embed_type": null,
|
||||
"addition_embed_type_num_heads": 64,
|
||||
"addition_time_embed_dim": null,
|
||||
"attention_head_dim": 8,
|
||||
"block_out_channels": [
|
||||
320,
|
||||
640,
|
||||
1280,
|
||||
1280
|
||||
],
|
||||
"brushnet_conditioning_channel_order": "rgb",
|
||||
"class_embed_type": null,
|
||||
"conditioning_channels": 5,
|
||||
"conditioning_embedding_out_channels": [
|
||||
16,
|
||||
32,
|
||||
96,
|
||||
256
|
||||
],
|
||||
"cross_attention_dim": 768,
|
||||
"down_block_types": [
|
||||
"DownBlock2D",
|
||||
"DownBlock2D",
|
||||
"DownBlock2D",
|
||||
"DownBlock2D"
|
||||
],
|
||||
"downsample_padding": 1,
|
||||
"encoder_hid_dim": null,
|
||||
"encoder_hid_dim_type": null,
|
||||
"flip_sin_to_cos": true,
|
||||
"freq_shift": 0,
|
||||
"global_pool_conditions": false,
|
||||
"in_channels": 4,
|
||||
"layers_per_block": 2,
|
||||
"mid_block_scale_factor": 1,
|
||||
"mid_block_type": "MidBlock2D",
|
||||
"norm_eps": 1e-05,
|
||||
"norm_num_groups": 32,
|
||||
"num_attention_heads": null,
|
||||
"num_class_embeds": null,
|
||||
"only_cross_attention": false,
|
||||
"projection_class_embeddings_input_dim": null,
|
||||
"resnet_time_scale_shift": "default",
|
||||
"transformer_layers_per_block": 1,
|
||||
"up_block_types": [
|
||||
"UpBlock2D",
|
||||
"UpBlock2D",
|
||||
"UpBlock2D",
|
||||
"UpBlock2D"
|
||||
],
|
||||
"upcast_attention": false,
|
||||
"use_linear_projection": false
|
||||
}
|
||||
@@ -0,0 +1,932 @@
|
||||
from dataclasses import dataclass
|
||||
from typing import Any, Dict, List, Optional, Tuple, Union
|
||||
|
||||
import torch
|
||||
from torch import nn
|
||||
from torch.nn import functional as F
|
||||
|
||||
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
||||
from diffusers.utils import BaseOutput, logging
|
||||
from diffusers.models.attention_processor import (
|
||||
ADDED_KV_ATTENTION_PROCESSORS,
|
||||
CROSS_ATTENTION_PROCESSORS,
|
||||
AttentionProcessor,
|
||||
AttnAddedKVProcessor,
|
||||
AttnProcessor,
|
||||
)
|
||||
from diffusers.models.embeddings import TextImageProjection, TextImageTimeEmbedding, TextTimeEmbedding, TimestepEmbedding, Timesteps
|
||||
from diffusers.models.modeling_utils import ModelMixin
|
||||
|
||||
from .unet_2d_blocks import (
|
||||
CrossAttnDownBlock2D,
|
||||
DownBlock2D,
|
||||
UNetMidBlock2D,
|
||||
UNetMidBlock2DCrossAttn,
|
||||
get_down_block,
|
||||
get_mid_block,
|
||||
get_up_block,
|
||||
MidBlock2D
|
||||
)
|
||||
|
||||
from .unet_2d_condition import UNet2DConditionModel
|
||||
|
||||
|
||||
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
||||
|
||||
|
||||
@dataclass
|
||||
class BrushNetOutput(BaseOutput):
|
||||
"""
|
||||
The output of [`BrushNetModel`].
|
||||
|
||||
Args:
|
||||
up_block_res_samples (`tuple[torch.Tensor]`):
|
||||
A tuple of upsample activations at different resolutions for each upsampling block. Each tensor should
|
||||
be of shape `(batch_size, channel * resolution, height //resolution, width // resolution)`. Output can be
|
||||
used to condition the original UNet's upsampling activations.
|
||||
down_block_res_samples (`tuple[torch.Tensor]`):
|
||||
A tuple of downsample activations at different resolutions for each downsampling block. Each tensor should
|
||||
be of shape `(batch_size, channel * resolution, height //resolution, width // resolution)`. Output can be
|
||||
used to condition the original UNet's downsampling activations.
|
||||
mid_down_block_re_sample (`torch.Tensor`):
|
||||
The activation of the midde block (the lowest sample resolution). Each tensor should be of shape
|
||||
`(batch_size, channel * lowest_resolution, height // lowest_resolution, width // lowest_resolution)`.
|
||||
Output can be used to condition the original UNet's middle block activation.
|
||||
"""
|
||||
|
||||
up_block_res_samples: Tuple[torch.Tensor]
|
||||
down_block_res_samples: Tuple[torch.Tensor]
|
||||
mid_block_res_sample: torch.Tensor
|
||||
|
||||
|
||||
class BrushNetModel(ModelMixin, ConfigMixin):
|
||||
"""
|
||||
A BrushNet model.
|
||||
|
||||
Args:
|
||||
in_channels (`int`, defaults to 4):
|
||||
The number of channels in the input sample.
|
||||
flip_sin_to_cos (`bool`, defaults to `True`):
|
||||
Whether to flip the sin to cos in the time embedding.
|
||||
freq_shift (`int`, defaults to 0):
|
||||
The frequency shift to apply to the time embedding.
|
||||
down_block_types (`tuple[str]`, defaults to `("CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "DownBlock2D")`):
|
||||
The tuple of downsample blocks to use.
|
||||
mid_block_type (`str`, *optional*, defaults to `"UNetMidBlock2DCrossAttn"`):
|
||||
Block type for middle of UNet, it can be one of `UNetMidBlock2DCrossAttn`, `UNetMidBlock2D`, or
|
||||
`UNetMidBlock2DSimpleCrossAttn`. If `None`, the mid block layer is skipped.
|
||||
up_block_types (`Tuple[str]`, *optional*, defaults to `("UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D")`):
|
||||
The tuple of upsample blocks to use.
|
||||
only_cross_attention (`Union[bool, Tuple[bool]]`, defaults to `False`):
|
||||
block_out_channels (`tuple[int]`, defaults to `(320, 640, 1280, 1280)`):
|
||||
The tuple of output channels for each block.
|
||||
layers_per_block (`int`, defaults to 2):
|
||||
The number of layers per block.
|
||||
downsample_padding (`int`, defaults to 1):
|
||||
The padding to use for the downsampling convolution.
|
||||
mid_block_scale_factor (`float`, defaults to 1):
|
||||
The scale factor to use for the mid block.
|
||||
act_fn (`str`, defaults to "silu"):
|
||||
The activation function to use.
|
||||
norm_num_groups (`int`, *optional*, defaults to 32):
|
||||
The number of groups to use for the normalization. If None, normalization and activation layers is skipped
|
||||
in post-processing.
|
||||
norm_eps (`float`, defaults to 1e-5):
|
||||
The epsilon to use for the normalization.
|
||||
cross_attention_dim (`int`, defaults to 1280):
|
||||
The dimension of the cross attention features.
|
||||
transformer_layers_per_block (`int` or `Tuple[int]`, *optional*, defaults to 1):
|
||||
The number of transformer blocks of type [`~models.attention.BasicTransformerBlock`]. Only relevant for
|
||||
[`~models.unet_2d_blocks.CrossAttnDownBlock2D`], [`~models.unet_2d_blocks.CrossAttnUpBlock2D`],
|
||||
[`~models.unet_2d_blocks.UNetMidBlock2DCrossAttn`].
|
||||
encoder_hid_dim (`int`, *optional*, defaults to None):
|
||||
If `encoder_hid_dim_type` is defined, `encoder_hidden_states` will be projected from `encoder_hid_dim`
|
||||
dimension to `cross_attention_dim`.
|
||||
encoder_hid_dim_type (`str`, *optional*, defaults to `None`):
|
||||
If given, the `encoder_hidden_states` and potentially other embeddings are down-projected to text
|
||||
embeddings of dimension `cross_attention` according to `encoder_hid_dim_type`.
|
||||
attention_head_dim (`Union[int, Tuple[int]]`, defaults to 8):
|
||||
The dimension of the attention heads.
|
||||
use_linear_projection (`bool`, defaults to `False`):
|
||||
class_embed_type (`str`, *optional*, defaults to `None`):
|
||||
The type of class embedding to use which is ultimately summed with the time embeddings. Choose from None,
|
||||
`"timestep"`, `"identity"`, `"projection"`, or `"simple_projection"`.
|
||||
addition_embed_type (`str`, *optional*, defaults to `None`):
|
||||
Configures an optional embedding which will be summed with the time embeddings. Choose from `None` or
|
||||
"text". "text" will use the `TextTimeEmbedding` layer.
|
||||
num_class_embeds (`int`, *optional*, defaults to 0):
|
||||
Input dimension of the learnable embedding matrix to be projected to `time_embed_dim`, when performing
|
||||
class conditioning with `class_embed_type` equal to `None`.
|
||||
upcast_attention (`bool`, defaults to `False`):
|
||||
resnet_time_scale_shift (`str`, defaults to `"default"`):
|
||||
Time scale shift config for ResNet blocks (see `ResnetBlock2D`). Choose from `default` or `scale_shift`.
|
||||
projection_class_embeddings_input_dim (`int`, *optional*, defaults to `None`):
|
||||
The dimension of the `class_labels` input when `class_embed_type="projection"`. Required when
|
||||
`class_embed_type="projection"`.
|
||||
brushnet_conditioning_channel_order (`str`, defaults to `"rgb"`):
|
||||
The channel order of conditional image. Will convert to `rgb` if it's `bgr`.
|
||||
conditioning_embedding_out_channels (`tuple[int]`, *optional*, defaults to `(16, 32, 96, 256)`):
|
||||
The tuple of output channel for each block in the `conditioning_embedding` layer.
|
||||
global_pool_conditions (`bool`, defaults to `False`):
|
||||
TODO(Patrick) - unused parameter.
|
||||
addition_embed_type_num_heads (`int`, defaults to 64):
|
||||
The number of heads to use for the `TextTimeEmbedding` layer.
|
||||
"""
|
||||
|
||||
_supports_gradient_checkpointing = True
|
||||
|
||||
@register_to_config
|
||||
def __init__(
|
||||
self,
|
||||
in_channels: int = 4,
|
||||
conditioning_channels: int = 5,
|
||||
flip_sin_to_cos: bool = True,
|
||||
freq_shift: int = 0,
|
||||
down_block_types: Tuple[str, ...] = (
|
||||
"DownBlock2D",
|
||||
"DownBlock2D",
|
||||
"DownBlock2D",
|
||||
"DownBlock2D",
|
||||
),
|
||||
mid_block_type: Optional[str] = "UNetMidBlock2D",
|
||||
up_block_types: Tuple[str, ...] = (
|
||||
"UpBlock2D",
|
||||
"UpBlock2D",
|
||||
"UpBlock2D",
|
||||
"UpBlock2D",
|
||||
),
|
||||
only_cross_attention: Union[bool, Tuple[bool]] = False,
|
||||
block_out_channels: Tuple[int, ...] = (320, 640, 1280, 1280),
|
||||
layers_per_block: int = 2,
|
||||
downsample_padding: int = 1,
|
||||
mid_block_scale_factor: float = 1,
|
||||
act_fn: str = "silu",
|
||||
norm_num_groups: Optional[int] = 32,
|
||||
norm_eps: float = 1e-5,
|
||||
cross_attention_dim: int = 1280,
|
||||
transformer_layers_per_block: Union[int, Tuple[int, ...]] = 1,
|
||||
encoder_hid_dim: Optional[int] = None,
|
||||
encoder_hid_dim_type: Optional[str] = None,
|
||||
attention_head_dim: Union[int, Tuple[int, ...]] = 8,
|
||||
num_attention_heads: Optional[Union[int, Tuple[int, ...]]] = None,
|
||||
use_linear_projection: bool = False,
|
||||
class_embed_type: Optional[str] = None,
|
||||
addition_embed_type: Optional[str] = None,
|
||||
addition_time_embed_dim: Optional[int] = None,
|
||||
num_class_embeds: Optional[int] = None,
|
||||
upcast_attention: bool = False,
|
||||
resnet_time_scale_shift: str = "default",
|
||||
projection_class_embeddings_input_dim: Optional[int] = None,
|
||||
brushnet_conditioning_channel_order: str = "rgb",
|
||||
conditioning_embedding_out_channels: Optional[Tuple[int, ...]] = (16, 32, 96, 256),
|
||||
global_pool_conditions: bool = False,
|
||||
addition_embed_type_num_heads: int = 64,
|
||||
):
|
||||
super().__init__()
|
||||
|
||||
# If `num_attention_heads` is not defined (which is the case for most models)
|
||||
# it will default to `attention_head_dim`. This looks weird upon first reading it and it is.
|
||||
# The reason for this behavior is to correct for incorrectly named variables that were introduced
|
||||
# when this library was created. The incorrect naming was only discovered much later in https://github.com/huggingface/diffusers/issues/2011#issuecomment-1547958131
|
||||
# Changing `attention_head_dim` to `num_attention_heads` for 40,000+ configurations is too backwards breaking
|
||||
# which is why we correct for the naming here.
|
||||
num_attention_heads = num_attention_heads or attention_head_dim
|
||||
|
||||
# Check inputs
|
||||
if len(down_block_types) != len(up_block_types):
|
||||
raise ValueError(
|
||||
f"Must provide the same number of `down_block_types` as `up_block_types`. `down_block_types`: {down_block_types}. `up_block_types`: {up_block_types}."
|
||||
)
|
||||
|
||||
if len(block_out_channels) != len(down_block_types):
|
||||
raise ValueError(
|
||||
f"Must provide the same number of `block_out_channels` as `down_block_types`. `block_out_channels`: {block_out_channels}. `down_block_types`: {down_block_types}."
|
||||
)
|
||||
|
||||
if not isinstance(only_cross_attention, bool) and len(only_cross_attention) != len(down_block_types):
|
||||
raise ValueError(
|
||||
f"Must provide the same number of `only_cross_attention` as `down_block_types`. `only_cross_attention`: {only_cross_attention}. `down_block_types`: {down_block_types}."
|
||||
)
|
||||
|
||||
if not isinstance(num_attention_heads, int) and len(num_attention_heads) != len(down_block_types):
|
||||
raise ValueError(
|
||||
f"Must provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`: {num_attention_heads}. `down_block_types`: {down_block_types}."
|
||||
)
|
||||
|
||||
if isinstance(transformer_layers_per_block, int):
|
||||
transformer_layers_per_block = [transformer_layers_per_block] * len(down_block_types)
|
||||
|
||||
# input
|
||||
conv_in_kernel = 3
|
||||
conv_in_padding = (conv_in_kernel - 1) // 2
|
||||
self.conv_in_condition = nn.Conv2d(
|
||||
in_channels+conditioning_channels, block_out_channels[0], kernel_size=conv_in_kernel, padding=conv_in_padding
|
||||
)
|
||||
|
||||
# time
|
||||
time_embed_dim = block_out_channels[0] * 4
|
||||
self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
|
||||
timestep_input_dim = block_out_channels[0]
|
||||
self.time_embedding = TimestepEmbedding(
|
||||
timestep_input_dim,
|
||||
time_embed_dim,
|
||||
act_fn=act_fn,
|
||||
)
|
||||
|
||||
if encoder_hid_dim_type is None and encoder_hid_dim is not None:
|
||||
encoder_hid_dim_type = "text_proj"
|
||||
self.register_to_config(encoder_hid_dim_type=encoder_hid_dim_type)
|
||||
logger.info("encoder_hid_dim_type defaults to 'text_proj' as `encoder_hid_dim` is defined.")
|
||||
|
||||
if encoder_hid_dim is None and encoder_hid_dim_type is not None:
|
||||
raise ValueError(
|
||||
f"`encoder_hid_dim` has to be defined when `encoder_hid_dim_type` is set to {encoder_hid_dim_type}."
|
||||
)
|
||||
|
||||
if encoder_hid_dim_type == "text_proj":
|
||||
self.encoder_hid_proj = nn.Linear(encoder_hid_dim, cross_attention_dim)
|
||||
elif encoder_hid_dim_type == "text_image_proj":
|
||||
# image_embed_dim DOESN'T have to be `cross_attention_dim`. To not clutter the __init__ too much
|
||||
# they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
|
||||
# case when `addition_embed_type == "text_image_proj"` (Kadinsky 2.1)`
|
||||
self.encoder_hid_proj = TextImageProjection(
|
||||
text_embed_dim=encoder_hid_dim,
|
||||
image_embed_dim=cross_attention_dim,
|
||||
cross_attention_dim=cross_attention_dim,
|
||||
)
|
||||
|
||||
elif encoder_hid_dim_type is not None:
|
||||
raise ValueError(
|
||||
f"encoder_hid_dim_type: {encoder_hid_dim_type} must be None, 'text_proj' or 'text_image_proj'."
|
||||
)
|
||||
else:
|
||||
self.encoder_hid_proj = None
|
||||
|
||||
# class embedding
|
||||
if class_embed_type is None and num_class_embeds is not None:
|
||||
self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
|
||||
elif class_embed_type == "timestep":
|
||||
self.class_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
|
||||
elif class_embed_type == "identity":
|
||||
self.class_embedding = nn.Identity(time_embed_dim, time_embed_dim)
|
||||
elif class_embed_type == "projection":
|
||||
if projection_class_embeddings_input_dim is None:
|
||||
raise ValueError(
|
||||
"`class_embed_type`: 'projection' requires `projection_class_embeddings_input_dim` be set"
|
||||
)
|
||||
# The projection `class_embed_type` is the same as the timestep `class_embed_type` except
|
||||
# 1. the `class_labels` inputs are not first converted to sinusoidal embeddings
|
||||
# 2. it projects from an arbitrary input dimension.
|
||||
#
|
||||
# Note that `TimestepEmbedding` is quite general, being mainly linear layers and activations.
|
||||
# When used for embedding actual timesteps, the timesteps are first converted to sinusoidal embeddings.
|
||||
# As a result, `TimestepEmbedding` can be passed arbitrary vectors.
|
||||
self.class_embedding = TimestepEmbedding(projection_class_embeddings_input_dim, time_embed_dim)
|
||||
else:
|
||||
self.class_embedding = None
|
||||
|
||||
if addition_embed_type == "text":
|
||||
if encoder_hid_dim is not None:
|
||||
text_time_embedding_from_dim = encoder_hid_dim
|
||||
else:
|
||||
text_time_embedding_from_dim = cross_attention_dim
|
||||
|
||||
self.add_embedding = TextTimeEmbedding(
|
||||
text_time_embedding_from_dim, time_embed_dim, num_heads=addition_embed_type_num_heads
|
||||
)
|
||||
elif addition_embed_type == "text_image":
|
||||
# text_embed_dim and image_embed_dim DON'T have to be `cross_attention_dim`. To not clutter the __init__ too much
|
||||
# they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
|
||||
# case when `addition_embed_type == "text_image"` (Kadinsky 2.1)`
|
||||
self.add_embedding = TextImageTimeEmbedding(
|
||||
text_embed_dim=cross_attention_dim, image_embed_dim=cross_attention_dim, time_embed_dim=time_embed_dim
|
||||
)
|
||||
elif addition_embed_type == "text_time":
|
||||
self.add_time_proj = Timesteps(addition_time_embed_dim, flip_sin_to_cos, freq_shift)
|
||||
self.add_embedding = TimestepEmbedding(projection_class_embeddings_input_dim, time_embed_dim)
|
||||
|
||||
elif addition_embed_type is not None:
|
||||
raise ValueError(f"addition_embed_type: {addition_embed_type} must be None, 'text' or 'text_image'.")
|
||||
|
||||
self.down_blocks = nn.ModuleList([])
|
||||
self.brushnet_down_blocks = nn.ModuleList([])
|
||||
|
||||
if isinstance(only_cross_attention, bool):
|
||||
only_cross_attention = [only_cross_attention] * len(down_block_types)
|
||||
|
||||
if isinstance(attention_head_dim, int):
|
||||
attention_head_dim = (attention_head_dim,) * len(down_block_types)
|
||||
|
||||
if isinstance(num_attention_heads, int):
|
||||
num_attention_heads = (num_attention_heads,) * len(down_block_types)
|
||||
|
||||
# down
|
||||
output_channel = block_out_channels[0]
|
||||
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_down_blocks.append(brushnet_block)
|
||||
|
||||
for i, down_block_type in enumerate(down_block_types):
|
||||
input_channel = output_channel
|
||||
output_channel = block_out_channels[i]
|
||||
is_final_block = i == len(block_out_channels) - 1
|
||||
|
||||
down_block = get_down_block(
|
||||
down_block_type,
|
||||
num_layers=layers_per_block,
|
||||
transformer_layers_per_block=transformer_layers_per_block[i],
|
||||
in_channels=input_channel,
|
||||
out_channels=output_channel,
|
||||
temb_channels=time_embed_dim,
|
||||
add_downsample=not is_final_block,
|
||||
resnet_eps=norm_eps,
|
||||
resnet_act_fn=act_fn,
|
||||
resnet_groups=norm_num_groups,
|
||||
cross_attention_dim=cross_attention_dim,
|
||||
num_attention_heads=num_attention_heads[i],
|
||||
attention_head_dim=attention_head_dim[i] if attention_head_dim[i] is not None else output_channel,
|
||||
downsample_padding=downsample_padding,
|
||||
use_linear_projection=use_linear_projection,
|
||||
only_cross_attention=only_cross_attention[i],
|
||||
upcast_attention=upcast_attention,
|
||||
resnet_time_scale_shift=resnet_time_scale_shift,
|
||||
)
|
||||
self.down_blocks.append(down_block)
|
||||
|
||||
for _ in range(layers_per_block):
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_down_blocks.append(brushnet_block)
|
||||
|
||||
if not is_final_block:
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_down_blocks.append(brushnet_block)
|
||||
|
||||
# mid
|
||||
mid_block_channel = block_out_channels[-1]
|
||||
|
||||
brushnet_block = nn.Conv2d(mid_block_channel, mid_block_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_mid_block = brushnet_block
|
||||
|
||||
self.mid_block = get_mid_block(
|
||||
mid_block_type,
|
||||
transformer_layers_per_block=transformer_layers_per_block[-1],
|
||||
in_channels=mid_block_channel,
|
||||
temb_channels=time_embed_dim,
|
||||
resnet_eps=norm_eps,
|
||||
resnet_act_fn=act_fn,
|
||||
output_scale_factor=mid_block_scale_factor,
|
||||
resnet_time_scale_shift=resnet_time_scale_shift,
|
||||
cross_attention_dim=cross_attention_dim,
|
||||
num_attention_heads=num_attention_heads[-1],
|
||||
resnet_groups=norm_num_groups,
|
||||
use_linear_projection=use_linear_projection,
|
||||
upcast_attention=upcast_attention,
|
||||
)
|
||||
|
||||
# count how many layers upsample the images
|
||||
self.num_upsamplers = 0
|
||||
|
||||
# up
|
||||
reversed_block_out_channels = list(reversed(block_out_channels))
|
||||
reversed_num_attention_heads = list(reversed(num_attention_heads))
|
||||
reversed_transformer_layers_per_block = (list(reversed(transformer_layers_per_block)))
|
||||
only_cross_attention = list(reversed(only_cross_attention))
|
||||
|
||||
output_channel = reversed_block_out_channels[0]
|
||||
|
||||
self.up_blocks = nn.ModuleList([])
|
||||
self.brushnet_up_blocks = nn.ModuleList([])
|
||||
|
||||
for i, up_block_type in enumerate(up_block_types):
|
||||
is_final_block = i == len(block_out_channels) - 1
|
||||
|
||||
prev_output_channel = output_channel
|
||||
output_channel = reversed_block_out_channels[i]
|
||||
input_channel = reversed_block_out_channels[min(i + 1, len(block_out_channels) - 1)]
|
||||
|
||||
# add upsample block for all BUT final layer
|
||||
if not is_final_block:
|
||||
add_upsample = True
|
||||
self.num_upsamplers += 1
|
||||
else:
|
||||
add_upsample = False
|
||||
|
||||
up_block = get_up_block(
|
||||
up_block_type,
|
||||
num_layers=layers_per_block+1,
|
||||
transformer_layers_per_block=reversed_transformer_layers_per_block[i],
|
||||
in_channels=input_channel,
|
||||
out_channels=output_channel,
|
||||
prev_output_channel=prev_output_channel,
|
||||
temb_channels=time_embed_dim,
|
||||
add_upsample=add_upsample,
|
||||
resnet_eps=norm_eps,
|
||||
resnet_act_fn=act_fn,
|
||||
resolution_idx=i,
|
||||
resnet_groups=norm_num_groups,
|
||||
cross_attention_dim=cross_attention_dim,
|
||||
num_attention_heads=reversed_num_attention_heads[i],
|
||||
use_linear_projection=use_linear_projection,
|
||||
only_cross_attention=only_cross_attention[i],
|
||||
upcast_attention=upcast_attention,
|
||||
resnet_time_scale_shift=resnet_time_scale_shift,
|
||||
attention_head_dim=attention_head_dim[i] if attention_head_dim[i] is not None else output_channel,
|
||||
)
|
||||
self.up_blocks.append(up_block)
|
||||
prev_output_channel = output_channel
|
||||
|
||||
for _ in range(layers_per_block+1):
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_up_blocks.append(brushnet_block)
|
||||
|
||||
if not is_final_block:
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_up_blocks.append(brushnet_block)
|
||||
|
||||
|
||||
@classmethod
|
||||
def from_unet(
|
||||
cls,
|
||||
unet: UNet2DConditionModel,
|
||||
brushnet_conditioning_channel_order: str = "rgb",
|
||||
conditioning_embedding_out_channels: Optional[Tuple[int, ...]] = (16, 32, 96, 256),
|
||||
load_weights_from_unet: bool = True,
|
||||
conditioning_channels: int = 5,
|
||||
):
|
||||
r"""
|
||||
Instantiate a [`BrushNetModel`] from [`UNet2DConditionModel`].
|
||||
|
||||
Parameters:
|
||||
unet (`UNet2DConditionModel`):
|
||||
The UNet model weights to copy to the [`BrushNetModel`]. All configuration options are also copied
|
||||
where applicable.
|
||||
"""
|
||||
transformer_layers_per_block = (
|
||||
unet.config.transformer_layers_per_block if "transformer_layers_per_block" in unet.config else 1
|
||||
)
|
||||
encoder_hid_dim = unet.config.encoder_hid_dim if "encoder_hid_dim" in unet.config else None
|
||||
encoder_hid_dim_type = unet.config.encoder_hid_dim_type if "encoder_hid_dim_type" in unet.config else None
|
||||
addition_embed_type = unet.config.addition_embed_type if "addition_embed_type" in unet.config else None
|
||||
addition_time_embed_dim = (
|
||||
unet.config.addition_time_embed_dim if "addition_time_embed_dim" in unet.config else None
|
||||
)
|
||||
|
||||
brushnet = cls(
|
||||
in_channels=unet.config.in_channels,
|
||||
conditioning_channels=conditioning_channels,
|
||||
flip_sin_to_cos=unet.config.flip_sin_to_cos,
|
||||
freq_shift=unet.config.freq_shift,
|
||||
down_block_types=['DownBlock2D','DownBlock2D','DownBlock2D','DownBlock2D'],
|
||||
mid_block_type='MidBlock2D',
|
||||
up_block_types=['UpBlock2D','UpBlock2D','UpBlock2D','UpBlock2D'],
|
||||
only_cross_attention=unet.config.only_cross_attention,
|
||||
block_out_channels=unet.config.block_out_channels,
|
||||
layers_per_block=unet.config.layers_per_block,
|
||||
downsample_padding=unet.config.downsample_padding,
|
||||
mid_block_scale_factor=unet.config.mid_block_scale_factor,
|
||||
act_fn=unet.config.act_fn,
|
||||
norm_num_groups=unet.config.norm_num_groups,
|
||||
norm_eps=unet.config.norm_eps,
|
||||
cross_attention_dim=unet.config.cross_attention_dim,
|
||||
transformer_layers_per_block=transformer_layers_per_block,
|
||||
encoder_hid_dim=encoder_hid_dim,
|
||||
encoder_hid_dim_type=encoder_hid_dim_type,
|
||||
attention_head_dim=unet.config.attention_head_dim,
|
||||
num_attention_heads=unet.config.num_attention_heads,
|
||||
use_linear_projection=unet.config.use_linear_projection,
|
||||
class_embed_type=unet.config.class_embed_type,
|
||||
addition_embed_type=addition_embed_type,
|
||||
addition_time_embed_dim=addition_time_embed_dim,
|
||||
num_class_embeds=unet.config.num_class_embeds,
|
||||
upcast_attention=unet.config.upcast_attention,
|
||||
resnet_time_scale_shift=unet.config.resnet_time_scale_shift,
|
||||
projection_class_embeddings_input_dim=unet.config.projection_class_embeddings_input_dim,
|
||||
brushnet_conditioning_channel_order=brushnet_conditioning_channel_order,
|
||||
conditioning_embedding_out_channels=conditioning_embedding_out_channels,
|
||||
)
|
||||
|
||||
if load_weights_from_unet:
|
||||
conv_in_condition_weight=torch.zeros_like(brushnet.conv_in_condition.weight)
|
||||
conv_in_condition_weight[:,:4,...]=unet.conv_in.weight
|
||||
conv_in_condition_weight[:,4:8,...]=unet.conv_in.weight
|
||||
brushnet.conv_in_condition.weight=torch.nn.Parameter(conv_in_condition_weight)
|
||||
brushnet.conv_in_condition.bias=unet.conv_in.bias
|
||||
|
||||
brushnet.time_proj.load_state_dict(unet.time_proj.state_dict())
|
||||
brushnet.time_embedding.load_state_dict(unet.time_embedding.state_dict())
|
||||
|
||||
if brushnet.class_embedding:
|
||||
brushnet.class_embedding.load_state_dict(unet.class_embedding.state_dict())
|
||||
|
||||
brushnet.down_blocks.load_state_dict(unet.down_blocks.state_dict(),strict=False)
|
||||
brushnet.mid_block.load_state_dict(unet.mid_block.state_dict(),strict=False)
|
||||
brushnet.up_blocks.load_state_dict(unet.up_blocks.state_dict(),strict=False)
|
||||
|
||||
return brushnet
|
||||
|
||||
@property
|
||||
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.attn_processors
|
||||
def attn_processors(self) -> Dict[str, AttentionProcessor]:
|
||||
r"""
|
||||
Returns:
|
||||
`dict` of attention processors: A dictionary containing all attention processors used in the model with
|
||||
indexed by its weight name.
|
||||
"""
|
||||
# set recursively
|
||||
processors = {}
|
||||
|
||||
def fn_recursive_add_processors(name: str, module: torch.nn.Module, processors: Dict[str, AttentionProcessor]):
|
||||
if hasattr(module, "get_processor"):
|
||||
processors[f"{name}.processor"] = module.get_processor(return_deprecated_lora=True)
|
||||
|
||||
for sub_name, child in module.named_children():
|
||||
fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)
|
||||
|
||||
return processors
|
||||
|
||||
for name, module in self.named_children():
|
||||
fn_recursive_add_processors(name, module, processors)
|
||||
|
||||
return processors
|
||||
|
||||
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attn_processor
|
||||
def set_attn_processor(self, processor: Union[AttentionProcessor, Dict[str, AttentionProcessor]]):
|
||||
r"""
|
||||
Sets the attention processor to use to compute attention.
|
||||
|
||||
Parameters:
|
||||
processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`):
|
||||
The instantiated processor class or a dictionary of processor classes that will be set as the processor
|
||||
for **all** `Attention` layers.
|
||||
|
||||
If `processor` is a dict, the key needs to define the path to the corresponding cross attention
|
||||
processor. This is strongly recommended when setting trainable attention processors.
|
||||
|
||||
"""
|
||||
count = len(self.attn_processors.keys())
|
||||
|
||||
if isinstance(processor, dict) and len(processor) != count:
|
||||
raise ValueError(
|
||||
f"A dict of processors was passed, but the number of processors {len(processor)} does not match the"
|
||||
f" number of attention layers: {count}. Please make sure to pass {count} processor classes."
|
||||
)
|
||||
|
||||
def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor):
|
||||
if hasattr(module, "set_processor"):
|
||||
if not isinstance(processor, dict):
|
||||
module.set_processor(processor)
|
||||
else:
|
||||
module.set_processor(processor.pop(f"{name}.processor"))
|
||||
|
||||
for sub_name, child in module.named_children():
|
||||
fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor)
|
||||
|
||||
for name, module in self.named_children():
|
||||
fn_recursive_attn_processor(name, module, processor)
|
||||
|
||||
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_default_attn_processor
|
||||
def set_default_attn_processor(self):
|
||||
"""
|
||||
Disables custom attention processors and sets the default attention implementation.
|
||||
"""
|
||||
if all(proc.__class__ in ADDED_KV_ATTENTION_PROCESSORS for proc in self.attn_processors.values()):
|
||||
processor = AttnAddedKVProcessor()
|
||||
elif all(proc.__class__ in CROSS_ATTENTION_PROCESSORS for proc in self.attn_processors.values()):
|
||||
processor = AttnProcessor()
|
||||
else:
|
||||
raise ValueError(
|
||||
f"Cannot call `set_default_attn_processor` when attention processors are of type {next(iter(self.attn_processors.values()))}"
|
||||
)
|
||||
|
||||
self.set_attn_processor(processor)
|
||||
|
||||
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attention_slice
|
||||
def set_attention_slice(self, slice_size: Union[str, int, List[int]]) -> None:
|
||||
r"""
|
||||
Enable sliced attention computation.
|
||||
|
||||
When this option is enabled, the attention module splits the input tensor in slices to compute attention in
|
||||
several steps. This is useful for saving some memory in exchange for a small decrease in speed.
|
||||
|
||||
Args:
|
||||
slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`):
|
||||
When `"auto"`, input to the attention heads is halved, so attention is computed in two steps. If
|
||||
`"max"`, maximum amount of memory is saved by running only one slice at a time. If a number is
|
||||
provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim`
|
||||
must be a multiple of `slice_size`.
|
||||
"""
|
||||
sliceable_head_dims = []
|
||||
|
||||
def fn_recursive_retrieve_sliceable_dims(module: torch.nn.Module):
|
||||
if hasattr(module, "set_attention_slice"):
|
||||
sliceable_head_dims.append(module.sliceable_head_dim)
|
||||
|
||||
for child in module.children():
|
||||
fn_recursive_retrieve_sliceable_dims(child)
|
||||
|
||||
# retrieve number of attention layers
|
||||
for module in self.children():
|
||||
fn_recursive_retrieve_sliceable_dims(module)
|
||||
|
||||
num_sliceable_layers = len(sliceable_head_dims)
|
||||
|
||||
if slice_size == "auto":
|
||||
# half the attention head size is usually a good trade-off between
|
||||
# speed and memory
|
||||
slice_size = [dim // 2 for dim in sliceable_head_dims]
|
||||
elif slice_size == "max":
|
||||
# make smallest slice possible
|
||||
slice_size = num_sliceable_layers * [1]
|
||||
|
||||
slice_size = num_sliceable_layers * [slice_size] if not isinstance(slice_size, list) else slice_size
|
||||
|
||||
if len(slice_size) != len(sliceable_head_dims):
|
||||
raise ValueError(
|
||||
f"You have provided {len(slice_size)}, but {self.config} has {len(sliceable_head_dims)} different"
|
||||
f" attention layers. Make sure to match `len(slice_size)` to be {len(sliceable_head_dims)}."
|
||||
)
|
||||
|
||||
for i in range(len(slice_size)):
|
||||
size = slice_size[i]
|
||||
dim = sliceable_head_dims[i]
|
||||
if size is not None and size > dim:
|
||||
raise ValueError(f"size {size} has to be smaller or equal to {dim}.")
|
||||
|
||||
# Recursively walk through all the children.
|
||||
# Any children which exposes the set_attention_slice method
|
||||
# gets the message
|
||||
def fn_recursive_set_attention_slice(module: torch.nn.Module, slice_size: List[int]):
|
||||
if hasattr(module, "set_attention_slice"):
|
||||
module.set_attention_slice(slice_size.pop())
|
||||
|
||||
for child in module.children():
|
||||
fn_recursive_set_attention_slice(child, slice_size)
|
||||
|
||||
reversed_slice_size = list(reversed(slice_size))
|
||||
for module in self.children():
|
||||
fn_recursive_set_attention_slice(module, reversed_slice_size)
|
||||
|
||||
def _set_gradient_checkpointing(self, module, value: bool = False) -> None:
|
||||
if isinstance(module, (CrossAttnDownBlock2D, DownBlock2D)):
|
||||
module.gradient_checkpointing = value
|
||||
|
||||
def forward(
|
||||
self,
|
||||
sample: torch.FloatTensor,
|
||||
timestep: Union[torch.Tensor, float, int],
|
||||
encoder_hidden_states: torch.Tensor,
|
||||
brushnet_cond: torch.FloatTensor,
|
||||
conditioning_scale: float = 1.0,
|
||||
class_labels: Optional[torch.Tensor] = None,
|
||||
timestep_cond: Optional[torch.Tensor] = None,
|
||||
attention_mask: Optional[torch.Tensor] = None,
|
||||
added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
|
||||
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
||||
guess_mode: bool = False,
|
||||
return_dict: bool = True,
|
||||
) -> Union[BrushNetOutput, Tuple[Tuple[torch.FloatTensor, ...], torch.FloatTensor]]:
|
||||
"""
|
||||
The [`BrushNetModel`] forward method.
|
||||
|
||||
Args:
|
||||
sample (`torch.FloatTensor`):
|
||||
The noisy input tensor.
|
||||
timestep (`Union[torch.Tensor, float, int]`):
|
||||
The number of timesteps to denoise an input.
|
||||
encoder_hidden_states (`torch.Tensor`):
|
||||
The encoder hidden states.
|
||||
brushnet_cond (`torch.FloatTensor`):
|
||||
The conditional input tensor of shape `(batch_size, sequence_length, hidden_size)`.
|
||||
conditioning_scale (`float`, defaults to `1.0`):
|
||||
The scale factor for BrushNet outputs.
|
||||
class_labels (`torch.Tensor`, *optional*, defaults to `None`):
|
||||
Optional class labels for conditioning. Their embeddings will be summed with the timestep embeddings.
|
||||
timestep_cond (`torch.Tensor`, *optional*, defaults to `None`):
|
||||
Additional conditional embeddings for timestep. If provided, the embeddings will be summed with the
|
||||
timestep_embedding passed through the `self.time_embedding` layer to obtain the final timestep
|
||||
embeddings.
|
||||
attention_mask (`torch.Tensor`, *optional*, defaults to `None`):
|
||||
An attention mask of shape `(batch, key_tokens)` is applied to `encoder_hidden_states`. If `1` the mask
|
||||
is kept, otherwise if `0` it is discarded. Mask will be converted into a bias, which adds large
|
||||
negative values to the attention scores corresponding to "discard" tokens.
|
||||
added_cond_kwargs (`dict`):
|
||||
Additional conditions for the Stable Diffusion XL UNet.
|
||||
cross_attention_kwargs (`dict[str]`, *optional*, defaults to `None`):
|
||||
A kwargs dictionary that if specified is passed along to the `AttnProcessor`.
|
||||
guess_mode (`bool`, defaults to `False`):
|
||||
In this mode, the BrushNet encoder tries its best to recognize the input content of the input even if
|
||||
you remove all prompts. A `guidance_scale` between 3.0 and 5.0 is recommended.
|
||||
return_dict (`bool`, defaults to `True`):
|
||||
Whether or not to return a [`~models.brushnet.BrushNetOutput`] instead of a plain tuple.
|
||||
|
||||
Returns:
|
||||
[`~models.brushnet.BrushNetOutput`] **or** `tuple`:
|
||||
If `return_dict` is `True`, a [`~models.brushnet.BrushNetOutput`] is returned, otherwise a tuple is
|
||||
returned where the first element is the sample tensor.
|
||||
"""
|
||||
# check channel order
|
||||
channel_order = self.config.brushnet_conditioning_channel_order
|
||||
|
||||
if channel_order == "rgb":
|
||||
# in rgb order by default
|
||||
...
|
||||
elif channel_order == "bgr":
|
||||
brushnet_cond = torch.flip(brushnet_cond, dims=[1])
|
||||
else:
|
||||
raise ValueError(f"unknown `brushnet_conditioning_channel_order`: {channel_order}")
|
||||
|
||||
# prepare attention_mask
|
||||
if attention_mask is not None:
|
||||
attention_mask = (1 - attention_mask.to(sample.dtype)) * -10000.0
|
||||
attention_mask = attention_mask.unsqueeze(1)
|
||||
|
||||
# 1. time
|
||||
timesteps = timestep
|
||||
if not torch.is_tensor(timesteps):
|
||||
# TODO: this requires sync between CPU and GPU. So try to pass timesteps as tensors if you can
|
||||
# This would be a good case for the `match` statement (Python 3.10+)
|
||||
is_mps = sample.device.type == "mps"
|
||||
if isinstance(timestep, float):
|
||||
dtype = torch.float32 if is_mps else torch.float64
|
||||
else:
|
||||
dtype = torch.int32 if is_mps else torch.int64
|
||||
timesteps = torch.tensor([timesteps], dtype=dtype, device=sample.device)
|
||||
elif len(timesteps.shape) == 0:
|
||||
timesteps = timesteps[None].to(sample.device)
|
||||
|
||||
# broadcast to batch dimension in a way that's compatible with ONNX/Core ML
|
||||
timesteps = timesteps.expand(sample.shape[0])
|
||||
|
||||
t_emb = self.time_proj(timesteps)
|
||||
|
||||
# timesteps does not contain any weights and will always return f32 tensors
|
||||
# but time_embedding might actually be running in fp16. so we need to cast here.
|
||||
# there might be better ways to encapsulate this.
|
||||
t_emb = t_emb.to(dtype=sample.dtype)
|
||||
|
||||
emb = self.time_embedding(t_emb, timestep_cond)
|
||||
aug_emb = None
|
||||
|
||||
if self.class_embedding is not None:
|
||||
if class_labels is None:
|
||||
raise ValueError("class_labels should be provided when num_class_embeds > 0")
|
||||
|
||||
if self.config.class_embed_type == "timestep":
|
||||
class_labels = self.time_proj(class_labels)
|
||||
|
||||
class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
|
||||
emb = emb + class_emb
|
||||
|
||||
if self.config.addition_embed_type is not None:
|
||||
if self.config.addition_embed_type == "text":
|
||||
aug_emb = self.add_embedding(encoder_hidden_states)
|
||||
|
||||
elif self.config.addition_embed_type == "text_time":
|
||||
if "text_embeds" not in added_cond_kwargs:
|
||||
raise ValueError(
|
||||
f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `text_embeds` to be passed in `added_cond_kwargs`"
|
||||
)
|
||||
text_embeds = added_cond_kwargs.get("text_embeds")
|
||||
if "time_ids" not in added_cond_kwargs:
|
||||
raise ValueError(
|
||||
f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `time_ids` to be passed in `added_cond_kwargs`"
|
||||
)
|
||||
time_ids = added_cond_kwargs.get("time_ids")
|
||||
time_embeds = self.add_time_proj(time_ids.flatten())
|
||||
time_embeds = time_embeds.reshape((text_embeds.shape[0], -1))
|
||||
|
||||
add_embeds = torch.concat([text_embeds, time_embeds], dim=-1)
|
||||
add_embeds = add_embeds.to(emb.dtype)
|
||||
aug_emb = self.add_embedding(add_embeds)
|
||||
|
||||
emb = emb + aug_emb if aug_emb is not None else emb
|
||||
|
||||
# 2. pre-process
|
||||
brushnet_cond=torch.concat([sample,brushnet_cond],1)
|
||||
sample = self.conv_in_condition(brushnet_cond)
|
||||
|
||||
|
||||
# 3. down
|
||||
down_block_res_samples = (sample,)
|
||||
for downsample_block in self.down_blocks:
|
||||
if hasattr(downsample_block, "has_cross_attention") and downsample_block.has_cross_attention:
|
||||
sample, res_samples = downsample_block(
|
||||
hidden_states=sample,
|
||||
temb=emb,
|
||||
encoder_hidden_states=encoder_hidden_states,
|
||||
attention_mask=attention_mask,
|
||||
cross_attention_kwargs=cross_attention_kwargs,
|
||||
)
|
||||
else:
|
||||
sample, res_samples = downsample_block(hidden_states=sample, temb=emb)
|
||||
|
||||
down_block_res_samples += res_samples
|
||||
|
||||
# 4. PaintingNet down blocks
|
||||
brushnet_down_block_res_samples = ()
|
||||
for down_block_res_sample, brushnet_down_block in zip(down_block_res_samples, self.brushnet_down_blocks):
|
||||
down_block_res_sample = brushnet_down_block(down_block_res_sample)
|
||||
brushnet_down_block_res_samples = brushnet_down_block_res_samples + (down_block_res_sample,)
|
||||
|
||||
|
||||
# 5. mid
|
||||
if self.mid_block is not None:
|
||||
if hasattr(self.mid_block, "has_cross_attention") and self.mid_block.has_cross_attention:
|
||||
sample = self.mid_block(
|
||||
sample,
|
||||
emb,
|
||||
encoder_hidden_states=encoder_hidden_states,
|
||||
attention_mask=attention_mask,
|
||||
cross_attention_kwargs=cross_attention_kwargs,
|
||||
)
|
||||
else:
|
||||
sample = self.mid_block(sample, emb)
|
||||
|
||||
# 6. BrushNet mid blocks
|
||||
brushnet_mid_block_res_sample = self.brushnet_mid_block(sample)
|
||||
|
||||
|
||||
# 7. up
|
||||
up_block_res_samples = ()
|
||||
for i, upsample_block in enumerate(self.up_blocks):
|
||||
is_final_block = i == len(self.up_blocks) - 1
|
||||
|
||||
res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
|
||||
down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]
|
||||
|
||||
# if we have not reached the final block and need to forward the
|
||||
# upsample size, we do it here
|
||||
if not is_final_block:
|
||||
upsample_size = down_block_res_samples[-1].shape[2:]
|
||||
|
||||
if hasattr(upsample_block, "has_cross_attention") and upsample_block.has_cross_attention:
|
||||
sample, up_res_samples = upsample_block(
|
||||
hidden_states=sample,
|
||||
temb=emb,
|
||||
res_hidden_states_tuple=res_samples,
|
||||
encoder_hidden_states=encoder_hidden_states,
|
||||
cross_attention_kwargs=cross_attention_kwargs,
|
||||
upsample_size=upsample_size,
|
||||
attention_mask=attention_mask,
|
||||
return_res_samples=True
|
||||
)
|
||||
else:
|
||||
sample, up_res_samples = upsample_block(
|
||||
hidden_states=sample,
|
||||
temb=emb,
|
||||
res_hidden_states_tuple=res_samples,
|
||||
upsample_size=upsample_size,
|
||||
return_res_samples=True
|
||||
)
|
||||
|
||||
up_block_res_samples += up_res_samples
|
||||
|
||||
# 8. BrushNet up blocks
|
||||
brushnet_up_block_res_samples = ()
|
||||
for up_block_res_sample, brushnet_up_block in zip(up_block_res_samples, self.brushnet_up_blocks):
|
||||
up_block_res_sample = brushnet_up_block(up_block_res_sample)
|
||||
brushnet_up_block_res_samples = brushnet_up_block_res_samples + (up_block_res_sample,)
|
||||
|
||||
# 6. scaling
|
||||
if guess_mode and not self.config.global_pool_conditions:
|
||||
scales = torch.logspace(-1, 0, len(brushnet_down_block_res_samples) + 1 + len(brushnet_up_block_res_samples), device=sample.device) # 0.1 to 1.0
|
||||
scales = scales * conditioning_scale
|
||||
|
||||
brushnet_down_block_res_samples = [sample * scale for sample, scale in zip(brushnet_down_block_res_samples, scales[:len(brushnet_down_block_res_samples)])]
|
||||
brushnet_mid_block_res_sample = brushnet_mid_block_res_sample * scales[len(brushnet_down_block_res_samples)]
|
||||
brushnet_up_block_res_samples = [sample * scale for sample, scale in zip(brushnet_up_block_res_samples, scales[len(brushnet_down_block_res_samples)+1:])]
|
||||
else:
|
||||
brushnet_down_block_res_samples = [sample * conditioning_scale for sample in brushnet_down_block_res_samples]
|
||||
brushnet_mid_block_res_sample = brushnet_mid_block_res_sample * conditioning_scale
|
||||
brushnet_up_block_res_samples = [sample * conditioning_scale for sample in brushnet_up_block_res_samples]
|
||||
|
||||
|
||||
if self.config.global_pool_conditions:
|
||||
brushnet_down_block_res_samples = [
|
||||
torch.mean(sample, dim=(2, 3), keepdim=True) for sample in brushnet_down_block_res_samples
|
||||
]
|
||||
brushnet_mid_block_res_sample = torch.mean(brushnet_mid_block_res_sample, dim=(2, 3), keepdim=True)
|
||||
brushnet_up_block_res_samples = [
|
||||
torch.mean(sample, dim=(2, 3), keepdim=True) for sample in brushnet_up_block_res_samples
|
||||
]
|
||||
|
||||
if not return_dict:
|
||||
return (brushnet_down_block_res_samples, brushnet_mid_block_res_sample, brushnet_up_block_res_samples)
|
||||
|
||||
return BrushNetOutput(
|
||||
down_block_res_samples=brushnet_down_block_res_samples,
|
||||
mid_block_res_sample=brushnet_mid_block_res_sample,
|
||||
up_block_res_samples=brushnet_up_block_res_samples
|
||||
)
|
||||
|
||||
|
||||
def zero_module(module):
|
||||
for p in module.parameters():
|
||||
nn.init.zeros_(p)
|
||||
return module
|
||||
@@ -0,0 +1,942 @@
|
||||
from dataclasses import dataclass
|
||||
from typing import Any, Dict, List, Optional, Tuple, Union
|
||||
|
||||
import torch
|
||||
from torch import nn
|
||||
from torch.nn import functional as F
|
||||
|
||||
from ..configuration_utils import ConfigMixin, register_to_config
|
||||
from ..utils import BaseOutput, logging
|
||||
from .attention_processor import (
|
||||
ADDED_KV_ATTENTION_PROCESSORS,
|
||||
CROSS_ATTENTION_PROCESSORS,
|
||||
AttentionProcessor,
|
||||
AttnAddedKVProcessor,
|
||||
AttnProcessor,
|
||||
)
|
||||
from .embeddings import TextImageProjection, TextImageTimeEmbedding, TextTimeEmbedding, TimestepEmbedding, Timesteps
|
||||
from .modeling_utils import ModelMixin
|
||||
from .unets.unet_2d_blocks import (
|
||||
CrossAttnDownBlock2D,
|
||||
DownBlock2D,
|
||||
UNetMidBlock2D,
|
||||
UNetMidBlock2DCrossAttn,
|
||||
get_down_block,
|
||||
get_mid_block,
|
||||
get_up_block,
|
||||
MidBlock2D
|
||||
)
|
||||
|
||||
from .unets.unet_2d_condition import UNet2DConditionModel
|
||||
|
||||
|
||||
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
||||
|
||||
|
||||
@dataclass
|
||||
class BrushNetOutput(BaseOutput):
|
||||
"""
|
||||
The output of [`BrushNetCAModel`].
|
||||
|
||||
Args:
|
||||
up_block_res_samples (`tuple[torch.Tensor]`):
|
||||
A tuple of upsample activations at different resolutions for each upsampling block. Each tensor should
|
||||
be of shape `(batch_size, channel * resolution, height //resolution, width // resolution)`. Output can be
|
||||
used to condition the original UNet's upsampling activations.
|
||||
down_block_res_samples (`tuple[torch.Tensor]`):
|
||||
A tuple of downsample activations at different resolutions for each downsampling block. Each tensor should
|
||||
be of shape `(batch_size, channel * resolution, height //resolution, width // resolution)`. Output can be
|
||||
used to condition the original UNet's downsampling activations.
|
||||
mid_down_block_re_sample (`torch.Tensor`):
|
||||
The activation of the midde block (the lowest sample resolution). Each tensor should be of shape
|
||||
`(batch_size, channel * lowest_resolution, height // lowest_resolution, width // lowest_resolution)`.
|
||||
Output can be used to condition the original UNet's middle block activation.
|
||||
"""
|
||||
|
||||
up_block_res_samples: Tuple[torch.Tensor]
|
||||
down_block_res_samples: Tuple[torch.Tensor]
|
||||
mid_block_res_sample: torch.Tensor
|
||||
|
||||
|
||||
class BrushNetCAModel(ModelMixin, ConfigMixin):
|
||||
"""
|
||||
A BrushNetCA model.
|
||||
|
||||
Args:
|
||||
in_channels (`int`, defaults to 4):
|
||||
The number of channels in the input sample.
|
||||
flip_sin_to_cos (`bool`, defaults to `True`):
|
||||
Whether to flip the sin to cos in the time embedding.
|
||||
freq_shift (`int`, defaults to 0):
|
||||
The frequency shift to apply to the time embedding.
|
||||
down_block_types (`tuple[str]`, defaults to `("CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "DownBlock2D")`):
|
||||
The tuple of downsample blocks to use.
|
||||
mid_block_type (`str`, *optional*, defaults to `"UNetMidBlock2DCrossAttn"`):
|
||||
Block type for middle of UNet, it can be one of `UNetMidBlock2DCrossAttn`, `UNetMidBlock2D`, or
|
||||
`UNetMidBlock2DSimpleCrossAttn`. If `None`, the mid block layer is skipped.
|
||||
up_block_types (`Tuple[str]`, *optional*, defaults to `("UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D")`):
|
||||
The tuple of upsample blocks to use.
|
||||
only_cross_attention (`Union[bool, Tuple[bool]]`, defaults to `False`):
|
||||
block_out_channels (`tuple[int]`, defaults to `(320, 640, 1280, 1280)`):
|
||||
The tuple of output channels for each block.
|
||||
layers_per_block (`int`, defaults to 2):
|
||||
The number of layers per block.
|
||||
downsample_padding (`int`, defaults to 1):
|
||||
The padding to use for the downsampling convolution.
|
||||
mid_block_scale_factor (`float`, defaults to 1):
|
||||
The scale factor to use for the mid block.
|
||||
act_fn (`str`, defaults to "silu"):
|
||||
The activation function to use.
|
||||
norm_num_groups (`int`, *optional*, defaults to 32):
|
||||
The number of groups to use for the normalization. If None, normalization and activation layers is skipped
|
||||
in post-processing.
|
||||
norm_eps (`float`, defaults to 1e-5):
|
||||
The epsilon to use for the normalization.
|
||||
cross_attention_dim (`int`, defaults to 1280):
|
||||
The dimension of the cross attention features.
|
||||
transformer_layers_per_block (`int` or `Tuple[int]`, *optional*, defaults to 1):
|
||||
The number of transformer blocks of type [`~models.attention.BasicTransformerBlock`]. Only relevant for
|
||||
[`~models.unet_2d_blocks.CrossAttnDownBlock2D`], [`~models.unet_2d_blocks.CrossAttnUpBlock2D`],
|
||||
[`~models.unet_2d_blocks.UNetMidBlock2DCrossAttn`].
|
||||
encoder_hid_dim (`int`, *optional*, defaults to None):
|
||||
If `encoder_hid_dim_type` is defined, `encoder_hidden_states` will be projected from `encoder_hid_dim`
|
||||
dimension to `cross_attention_dim`.
|
||||
encoder_hid_dim_type (`str`, *optional*, defaults to `None`):
|
||||
If given, the `encoder_hidden_states` and potentially other embeddings are down-projected to text
|
||||
embeddings of dimension `cross_attention` according to `encoder_hid_dim_type`.
|
||||
attention_head_dim (`Union[int, Tuple[int]]`, defaults to 8):
|
||||
The dimension of the attention heads.
|
||||
use_linear_projection (`bool`, defaults to `False`):
|
||||
class_embed_type (`str`, *optional*, defaults to `None`):
|
||||
The type of class embedding to use which is ultimately summed with the time embeddings. Choose from None,
|
||||
`"timestep"`, `"identity"`, `"projection"`, or `"simple_projection"`.
|
||||
addition_embed_type (`str`, *optional*, defaults to `None`):
|
||||
Configures an optional embedding which will be summed with the time embeddings. Choose from `None` or
|
||||
"text". "text" will use the `TextTimeEmbedding` layer.
|
||||
num_class_embeds (`int`, *optional*, defaults to 0):
|
||||
Input dimension of the learnable embedding matrix to be projected to `time_embed_dim`, when performing
|
||||
class conditioning with `class_embed_type` equal to `None`.
|
||||
upcast_attention (`bool`, defaults to `False`):
|
||||
resnet_time_scale_shift (`str`, defaults to `"default"`):
|
||||
Time scale shift config for ResNet blocks (see `ResnetBlock2D`). Choose from `default` or `scale_shift`.
|
||||
projection_class_embeddings_input_dim (`int`, *optional*, defaults to `None`):
|
||||
The dimension of the `class_labels` input when `class_embed_type="projection"`. Required when
|
||||
`class_embed_type="projection"`.
|
||||
brushnet_conditioning_channel_order (`str`, defaults to `"rgb"`):
|
||||
The channel order of conditional image. Will convert to `rgb` if it's `bgr`.
|
||||
conditioning_embedding_out_channels (`tuple[int]`, *optional*, defaults to `(16, 32, 96, 256)`):
|
||||
The tuple of output channel for each block in the `conditioning_embedding` layer.
|
||||
global_pool_conditions (`bool`, defaults to `False`):
|
||||
TODO(Patrick) - unused parameter.
|
||||
addition_embed_type_num_heads (`int`, defaults to 64):
|
||||
The number of heads to use for the `TextTimeEmbedding` layer.
|
||||
"""
|
||||
|
||||
_supports_gradient_checkpointing = True
|
||||
|
||||
@register_to_config
|
||||
def __init__(
|
||||
self,
|
||||
in_channels: int = 4,
|
||||
conditioning_channels: int = 5,
|
||||
flip_sin_to_cos: bool = True,
|
||||
freq_shift: int = 0,
|
||||
down_block_types: Tuple[str, ...] = (
|
||||
# "CrossAttnDownBlock2D",
|
||||
# "DownBlock2D",
|
||||
# "DownBlock2D",
|
||||
# "DownBlock2D",
|
||||
"CrossAttnDownBlock2D",
|
||||
"CrossAttnDownBlock2D",
|
||||
"CrossAttnDownBlock2D",
|
||||
"DownBlock2D",
|
||||
),
|
||||
mid_block_type: Optional[str] = "UNetMidBlock2DCrossAttn", #"UNetMidBlock2D",
|
||||
up_block_types: Tuple[str, ...] = (
|
||||
# "UpBlock2D",
|
||||
# "UpBlock2D",
|
||||
# "UpBlock2D",
|
||||
# "CrossAttnUpBlock2D",
|
||||
"UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D"
|
||||
),
|
||||
only_cross_attention: Union[bool, Tuple[bool]] = False,
|
||||
block_out_channels: Tuple[int, ...] = (320, 640, 1280, 1280),
|
||||
layers_per_block: int = 2,
|
||||
downsample_padding: int = 1,
|
||||
mid_block_scale_factor: float = 1,
|
||||
act_fn: str = "silu",
|
||||
norm_num_groups: Optional[int] = 32,
|
||||
norm_eps: float = 1e-5,
|
||||
cross_attention_dim: int = 1280,
|
||||
transformer_layers_per_block: Union[int, Tuple[int, ...]] = 1,
|
||||
encoder_hid_dim: Optional[int] = None,
|
||||
encoder_hid_dim_type: Optional[str] = None,
|
||||
attention_head_dim: Union[int, Tuple[int, ...]] = 8,
|
||||
num_attention_heads: Optional[Union[int, Tuple[int, ...]]] = None,
|
||||
use_linear_projection: bool = False,
|
||||
class_embed_type: Optional[str] = None,
|
||||
addition_embed_type: Optional[str] = None,
|
||||
addition_time_embed_dim: Optional[int] = None,
|
||||
num_class_embeds: Optional[int] = None,
|
||||
upcast_attention: bool = False,
|
||||
resnet_time_scale_shift: str = "default",
|
||||
projection_class_embeddings_input_dim: Optional[int] = None,
|
||||
brushnet_conditioning_channel_order: str = "rgb",
|
||||
conditioning_embedding_out_channels: Optional[Tuple[int, ...]] = (16, 32, 96, 256),
|
||||
global_pool_conditions: bool = False,
|
||||
addition_embed_type_num_heads: int = 64,
|
||||
):
|
||||
super().__init__()
|
||||
|
||||
# If `num_attention_heads` is not defined (which is the case for most models)
|
||||
# it will default to `attention_head_dim`. This looks weird upon first reading it and it is.
|
||||
# The reason for this behavior is to correct for incorrectly named variables that were introduced
|
||||
# when this library was created. The incorrect naming was only discovered much later in https://github.com/huggingface/diffusers/issues/2011#issuecomment-1547958131
|
||||
# Changing `attention_head_dim` to `num_attention_heads` for 40,000+ configurations is too backwards breaking
|
||||
# which is why we correct for the naming here.
|
||||
num_attention_heads = num_attention_heads or attention_head_dim
|
||||
|
||||
# Check inputs
|
||||
if len(down_block_types) != len(up_block_types):
|
||||
raise ValueError(
|
||||
f"Must provide the same number of `down_block_types` as `up_block_types`. `down_block_types`: {down_block_types}. `up_block_types`: {up_block_types}."
|
||||
)
|
||||
|
||||
if len(block_out_channels) != len(down_block_types):
|
||||
raise ValueError(
|
||||
f"Must provide the same number of `block_out_channels` as `down_block_types`. `block_out_channels`: {block_out_channels}. `down_block_types`: {down_block_types}."
|
||||
)
|
||||
|
||||
if not isinstance(only_cross_attention, bool) and len(only_cross_attention) != len(down_block_types):
|
||||
raise ValueError(
|
||||
f"Must provide the same number of `only_cross_attention` as `down_block_types`. `only_cross_attention`: {only_cross_attention}. `down_block_types`: {down_block_types}."
|
||||
)
|
||||
|
||||
if not isinstance(num_attention_heads, int) and len(num_attention_heads) != len(down_block_types):
|
||||
raise ValueError(
|
||||
f"Must provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`: {num_attention_heads}. `down_block_types`: {down_block_types}."
|
||||
)
|
||||
|
||||
if isinstance(transformer_layers_per_block, int):
|
||||
transformer_layers_per_block = [transformer_layers_per_block] * len(down_block_types)
|
||||
|
||||
# input
|
||||
conv_in_kernel = 3
|
||||
conv_in_padding = (conv_in_kernel - 1) // 2
|
||||
self.conv_in_condition = nn.Conv2d(
|
||||
in_channels+conditioning_channels, block_out_channels[0], kernel_size=conv_in_kernel, padding=conv_in_padding
|
||||
)
|
||||
|
||||
# time
|
||||
time_embed_dim = block_out_channels[0] * 4
|
||||
self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
|
||||
timestep_input_dim = block_out_channels[0]
|
||||
self.time_embedding = TimestepEmbedding(
|
||||
timestep_input_dim,
|
||||
time_embed_dim,
|
||||
act_fn=act_fn,
|
||||
)
|
||||
|
||||
if encoder_hid_dim_type is None and encoder_hid_dim is not None:
|
||||
encoder_hid_dim_type = "text_proj"
|
||||
self.register_to_config(encoder_hid_dim_type=encoder_hid_dim_type)
|
||||
logger.info("encoder_hid_dim_type defaults to 'text_proj' as `encoder_hid_dim` is defined.")
|
||||
|
||||
if encoder_hid_dim is None and encoder_hid_dim_type is not None:
|
||||
raise ValueError(
|
||||
f"`encoder_hid_dim` has to be defined when `encoder_hid_dim_type` is set to {encoder_hid_dim_type}."
|
||||
)
|
||||
|
||||
if encoder_hid_dim_type == "text_proj":
|
||||
self.encoder_hid_proj = nn.Linear(encoder_hid_dim, cross_attention_dim)
|
||||
elif encoder_hid_dim_type == "text_image_proj":
|
||||
# image_embed_dim DOESN'T have to be `cross_attention_dim`. To not clutter the __init__ too much
|
||||
# they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
|
||||
# case when `addition_embed_type == "text_image_proj"` (Kadinsky 2.1)`
|
||||
self.encoder_hid_proj = TextImageProjection(
|
||||
text_embed_dim=encoder_hid_dim,
|
||||
image_embed_dim=cross_attention_dim,
|
||||
cross_attention_dim=cross_attention_dim,
|
||||
)
|
||||
|
||||
elif encoder_hid_dim_type is not None:
|
||||
raise ValueError(
|
||||
f"encoder_hid_dim_type: {encoder_hid_dim_type} must be None, 'text_proj' or 'text_image_proj'."
|
||||
)
|
||||
else:
|
||||
self.encoder_hid_proj = None
|
||||
|
||||
# class embedding
|
||||
if class_embed_type is None and num_class_embeds is not None:
|
||||
self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
|
||||
elif class_embed_type == "timestep":
|
||||
self.class_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
|
||||
elif class_embed_type == "identity":
|
||||
self.class_embedding = nn.Identity(time_embed_dim, time_embed_dim)
|
||||
elif class_embed_type == "projection":
|
||||
if projection_class_embeddings_input_dim is None:
|
||||
raise ValueError(
|
||||
"`class_embed_type`: 'projection' requires `projection_class_embeddings_input_dim` be set"
|
||||
)
|
||||
# The projection `class_embed_type` is the same as the timestep `class_embed_type` except
|
||||
# 1. the `class_labels` inputs are not first converted to sinusoidal embeddings
|
||||
# 2. it projects from an arbitrary input dimension.
|
||||
#
|
||||
# Note that `TimestepEmbedding` is quite general, being mainly linear layers and activations.
|
||||
# When used for embedding actual timesteps, the timesteps are first converted to sinusoidal embeddings.
|
||||
# As a result, `TimestepEmbedding` can be passed arbitrary vectors.
|
||||
self.class_embedding = TimestepEmbedding(projection_class_embeddings_input_dim, time_embed_dim)
|
||||
else:
|
||||
self.class_embedding = None
|
||||
|
||||
if addition_embed_type == "text":
|
||||
if encoder_hid_dim is not None:
|
||||
text_time_embedding_from_dim = encoder_hid_dim
|
||||
else:
|
||||
text_time_embedding_from_dim = cross_attention_dim
|
||||
|
||||
self.add_embedding = TextTimeEmbedding(
|
||||
text_time_embedding_from_dim, time_embed_dim, num_heads=addition_embed_type_num_heads
|
||||
)
|
||||
elif addition_embed_type == "text_image":
|
||||
# text_embed_dim and image_embed_dim DON'T have to be `cross_attention_dim`. To not clutter the __init__ too much
|
||||
# they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
|
||||
# case when `addition_embed_type == "text_image"` (Kadinsky 2.1)`
|
||||
self.add_embedding = TextImageTimeEmbedding(
|
||||
text_embed_dim=cross_attention_dim, image_embed_dim=cross_attention_dim, time_embed_dim=time_embed_dim
|
||||
)
|
||||
elif addition_embed_type == "text_time":
|
||||
self.add_time_proj = Timesteps(addition_time_embed_dim, flip_sin_to_cos, freq_shift)
|
||||
self.add_embedding = TimestepEmbedding(projection_class_embeddings_input_dim, time_embed_dim)
|
||||
|
||||
elif addition_embed_type is not None:
|
||||
raise ValueError(f"addition_embed_type: {addition_embed_type} must be None, 'text' or 'text_image'.")
|
||||
|
||||
self.down_blocks = nn.ModuleList([])
|
||||
self.brushnet_down_blocks = nn.ModuleList([])
|
||||
|
||||
if isinstance(only_cross_attention, bool):
|
||||
only_cross_attention = [only_cross_attention] * len(down_block_types)
|
||||
|
||||
if isinstance(attention_head_dim, int):
|
||||
attention_head_dim = (attention_head_dim,) * len(down_block_types)
|
||||
|
||||
if isinstance(num_attention_heads, int):
|
||||
num_attention_heads = (num_attention_heads,) * len(down_block_types)
|
||||
|
||||
# down
|
||||
output_channel = block_out_channels[0]
|
||||
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_down_blocks.append(brushnet_block)
|
||||
|
||||
for i, down_block_type in enumerate(down_block_types):
|
||||
input_channel = output_channel
|
||||
output_channel = block_out_channels[i]
|
||||
is_final_block = i == len(block_out_channels) - 1
|
||||
|
||||
down_block = get_down_block(
|
||||
down_block_type,
|
||||
num_layers=layers_per_block,
|
||||
transformer_layers_per_block=transformer_layers_per_block[i],
|
||||
in_channels=input_channel,
|
||||
out_channels=output_channel,
|
||||
temb_channels=time_embed_dim,
|
||||
add_downsample=not is_final_block,
|
||||
resnet_eps=norm_eps,
|
||||
resnet_act_fn=act_fn,
|
||||
resnet_groups=norm_num_groups,
|
||||
cross_attention_dim=cross_attention_dim,
|
||||
num_attention_heads=num_attention_heads[i],
|
||||
attention_head_dim=attention_head_dim[i] if attention_head_dim[i] is not None else output_channel,
|
||||
downsample_padding=downsample_padding,
|
||||
use_linear_projection=use_linear_projection,
|
||||
only_cross_attention=only_cross_attention[i],
|
||||
upcast_attention=upcast_attention,
|
||||
resnet_time_scale_shift=resnet_time_scale_shift,
|
||||
)
|
||||
self.down_blocks.append(down_block)
|
||||
|
||||
for _ in range(layers_per_block):
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_down_blocks.append(brushnet_block)
|
||||
|
||||
if not is_final_block:
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_down_blocks.append(brushnet_block)
|
||||
|
||||
# mid
|
||||
mid_block_channel = block_out_channels[-1]
|
||||
|
||||
brushnet_block = nn.Conv2d(mid_block_channel, mid_block_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_mid_block = brushnet_block
|
||||
|
||||
self.mid_block = get_mid_block(
|
||||
mid_block_type,
|
||||
transformer_layers_per_block=transformer_layers_per_block[-1],
|
||||
in_channels=mid_block_channel,
|
||||
temb_channels=time_embed_dim,
|
||||
resnet_eps=norm_eps,
|
||||
resnet_act_fn=act_fn,
|
||||
output_scale_factor=mid_block_scale_factor,
|
||||
resnet_time_scale_shift=resnet_time_scale_shift,
|
||||
cross_attention_dim=cross_attention_dim,
|
||||
num_attention_heads=num_attention_heads[-1],
|
||||
resnet_groups=norm_num_groups,
|
||||
use_linear_projection=use_linear_projection,
|
||||
upcast_attention=upcast_attention,
|
||||
)
|
||||
|
||||
# count how many layers upsample the images
|
||||
self.num_upsamplers = 0
|
||||
|
||||
# up
|
||||
reversed_block_out_channels = list(reversed(block_out_channels))
|
||||
reversed_num_attention_heads = list(reversed(num_attention_heads))
|
||||
reversed_transformer_layers_per_block = (list(reversed(transformer_layers_per_block)))
|
||||
only_cross_attention = list(reversed(only_cross_attention))
|
||||
|
||||
output_channel = reversed_block_out_channels[0]
|
||||
|
||||
self.up_blocks = nn.ModuleList([])
|
||||
self.brushnet_up_blocks = nn.ModuleList([])
|
||||
|
||||
for i, up_block_type in enumerate(up_block_types):
|
||||
is_final_block = i == len(block_out_channels) - 1
|
||||
|
||||
prev_output_channel = output_channel
|
||||
output_channel = reversed_block_out_channels[i]
|
||||
input_channel = reversed_block_out_channels[min(i + 1, len(block_out_channels) - 1)]
|
||||
|
||||
# add upsample block for all BUT final layer
|
||||
if not is_final_block:
|
||||
add_upsample = True
|
||||
self.num_upsamplers += 1
|
||||
else:
|
||||
add_upsample = False
|
||||
|
||||
up_block = get_up_block(
|
||||
up_block_type,
|
||||
num_layers=layers_per_block+1,
|
||||
transformer_layers_per_block=reversed_transformer_layers_per_block[i],
|
||||
in_channels=input_channel,
|
||||
out_channels=output_channel,
|
||||
prev_output_channel=prev_output_channel,
|
||||
temb_channels=time_embed_dim,
|
||||
add_upsample=add_upsample,
|
||||
resnet_eps=norm_eps,
|
||||
resnet_act_fn=act_fn,
|
||||
resolution_idx=i,
|
||||
resnet_groups=norm_num_groups,
|
||||
cross_attention_dim=cross_attention_dim,
|
||||
num_attention_heads=reversed_num_attention_heads[i],
|
||||
use_linear_projection=use_linear_projection,
|
||||
only_cross_attention=only_cross_attention[i],
|
||||
upcast_attention=upcast_attention,
|
||||
resnet_time_scale_shift=resnet_time_scale_shift,
|
||||
attention_head_dim=attention_head_dim[i] if attention_head_dim[i] is not None else output_channel,
|
||||
)
|
||||
self.up_blocks.append(up_block)
|
||||
prev_output_channel = output_channel
|
||||
|
||||
for _ in range(layers_per_block+1):
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_up_blocks.append(brushnet_block)
|
||||
|
||||
if not is_final_block:
|
||||
brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1)
|
||||
brushnet_block = zero_module(brushnet_block)
|
||||
self.brushnet_up_blocks.append(brushnet_block)
|
||||
|
||||
|
||||
@classmethod
|
||||
def from_unet(
|
||||
cls,
|
||||
unet: UNet2DConditionModel,
|
||||
brushnet_conditioning_channel_order: str = "rgb",
|
||||
conditioning_embedding_out_channels: Optional[Tuple[int, ...]] = (16, 32, 96, 256),
|
||||
load_weights_from_unet: bool = True,
|
||||
conditioning_channels: int = 5,
|
||||
):
|
||||
r"""
|
||||
Instantiate a [`BrushNetCAModel`] from [`UNet2DConditionModel`].
|
||||
|
||||
Parameters:
|
||||
unet (`UNet2DConditionModel`):
|
||||
The UNet model weights to copy to the [`BrushNetCAModel`]. All configuration options are also copied
|
||||
where applicable.
|
||||
"""
|
||||
transformer_layers_per_block = (
|
||||
unet.config.transformer_layers_per_block if "transformer_layers_per_block" in unet.config else 1
|
||||
)
|
||||
encoder_hid_dim = unet.config.encoder_hid_dim if "encoder_hid_dim" in unet.config else None
|
||||
encoder_hid_dim_type = unet.config.encoder_hid_dim_type if "encoder_hid_dim_type" in unet.config else None
|
||||
addition_embed_type = unet.config.addition_embed_type if "addition_embed_type" in unet.config else None
|
||||
addition_time_embed_dim = (
|
||||
unet.config.addition_time_embed_dim if "addition_time_embed_dim" in unet.config else None
|
||||
)
|
||||
|
||||
brushnet = cls(
|
||||
in_channels=unet.config.in_channels,
|
||||
conditioning_channels=conditioning_channels,
|
||||
flip_sin_to_cos=unet.config.flip_sin_to_cos,
|
||||
freq_shift=unet.config.freq_shift,
|
||||
# down_block_types=['CrossAttnDownBlock2D','DownBlock2D','DownBlock2D','DownBlock2D'],
|
||||
down_block_types=["CrossAttnDownBlock2D",
|
||||
"CrossAttnDownBlock2D",
|
||||
"CrossAttnDownBlock2D",
|
||||
"DownBlock2D",],
|
||||
mid_block_type='UNetMidBlock2DCrossAttn',
|
||||
# mid_block_type="UNetMidBlock2DCrossAttn",
|
||||
# up_block_types=['UpBlock2D','UpBlock2D','UpBlock2D','CrossAttnUpBlock2D'],
|
||||
up_block_types=["UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D"],
|
||||
only_cross_attention=unet.config.only_cross_attention,
|
||||
block_out_channels=unet.config.block_out_channels,
|
||||
layers_per_block=unet.config.layers_per_block,
|
||||
downsample_padding=unet.config.downsample_padding,
|
||||
mid_block_scale_factor=unet.config.mid_block_scale_factor,
|
||||
act_fn=unet.config.act_fn,
|
||||
norm_num_groups=unet.config.norm_num_groups,
|
||||
norm_eps=unet.config.norm_eps,
|
||||
cross_attention_dim=unet.config.cross_attention_dim,
|
||||
transformer_layers_per_block=transformer_layers_per_block,
|
||||
encoder_hid_dim=encoder_hid_dim,
|
||||
encoder_hid_dim_type=encoder_hid_dim_type,
|
||||
attention_head_dim=unet.config.attention_head_dim,
|
||||
num_attention_heads=unet.config.num_attention_heads,
|
||||
use_linear_projection=unet.config.use_linear_projection,
|
||||
class_embed_type=unet.config.class_embed_type,
|
||||
addition_embed_type=addition_embed_type,
|
||||
addition_time_embed_dim=addition_time_embed_dim,
|
||||
num_class_embeds=unet.config.num_class_embeds,
|
||||
upcast_attention=unet.config.upcast_attention,
|
||||
resnet_time_scale_shift=unet.config.resnet_time_scale_shift,
|
||||
projection_class_embeddings_input_dim=unet.config.projection_class_embeddings_input_dim,
|
||||
brushnet_conditioning_channel_order=brushnet_conditioning_channel_order,
|
||||
conditioning_embedding_out_channels=conditioning_embedding_out_channels,
|
||||
)
|
||||
|
||||
if load_weights_from_unet:
|
||||
conv_in_condition_weight=torch.zeros_like(brushnet.conv_in_condition.weight)
|
||||
conv_in_condition_weight[:,:4,...]=unet.conv_in.weight
|
||||
conv_in_condition_weight[:,4:8,...]=unet.conv_in.weight
|
||||
brushnet.conv_in_condition.weight=torch.nn.Parameter(conv_in_condition_weight)
|
||||
brushnet.conv_in_condition.bias=unet.conv_in.bias
|
||||
|
||||
brushnet.time_proj.load_state_dict(unet.time_proj.state_dict())
|
||||
brushnet.time_embedding.load_state_dict(unet.time_embedding.state_dict())
|
||||
|
||||
if brushnet.class_embedding:
|
||||
brushnet.class_embedding.load_state_dict(unet.class_embedding.state_dict())
|
||||
|
||||
brushnet.down_blocks.load_state_dict(unet.down_blocks.state_dict(),strict=False)
|
||||
brushnet.mid_block.load_state_dict(unet.mid_block.state_dict(),strict=False)
|
||||
brushnet.up_blocks.load_state_dict(unet.up_blocks.state_dict(),strict=False)
|
||||
|
||||
return brushnet.to(unet.dtype)
|
||||
|
||||
@property
|
||||
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.attn_processors
|
||||
def attn_processors(self) -> Dict[str, AttentionProcessor]:
|
||||
r"""
|
||||
Returns:
|
||||
`dict` of attention processors: A dictionary containing all attention processors used in the model with
|
||||
indexed by its weight name.
|
||||
"""
|
||||
# set recursively
|
||||
processors = {}
|
||||
|
||||
def fn_recursive_add_processors(name: str, module: torch.nn.Module, processors: Dict[str, AttentionProcessor]):
|
||||
if hasattr(module, "get_processor"):
|
||||
processors[f"{name}.processor"] = module.get_processor(return_deprecated_lora=True)
|
||||
|
||||
for sub_name, child in module.named_children():
|
||||
fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)
|
||||
|
||||
return processors
|
||||
|
||||
for name, module in self.named_children():
|
||||
fn_recursive_add_processors(name, module, processors)
|
||||
|
||||
return processors
|
||||
|
||||
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attn_processor
|
||||
def set_attn_processor(self, processor: Union[AttentionProcessor, Dict[str, AttentionProcessor]]):
|
||||
r"""
|
||||
Sets the attention processor to use to compute attention.
|
||||
|
||||
Parameters:
|
||||
processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`):
|
||||
The instantiated processor class or a dictionary of processor classes that will be set as the processor
|
||||
for **all** `Attention` layers.
|
||||
|
||||
If `processor` is a dict, the key needs to define the path to the corresponding cross attention
|
||||
processor. This is strongly recommended when setting trainable attention processors.
|
||||
|
||||
"""
|
||||
count = len(self.attn_processors.keys())
|
||||
|
||||
if isinstance(processor, dict) and len(processor) != count:
|
||||
raise ValueError(
|
||||
f"A dict of processors was passed, but the number of processors {len(processor)} does not match the"
|
||||
f" number of attention layers: {count}. Please make sure to pass {count} processor classes."
|
||||
)
|
||||
|
||||
def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor):
|
||||
if hasattr(module, "set_processor"):
|
||||
if not isinstance(processor, dict):
|
||||
module.set_processor(processor)
|
||||
else:
|
||||
module.set_processor(processor.pop(f"{name}.processor"))
|
||||
|
||||
for sub_name, child in module.named_children():
|
||||
fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor)
|
||||
|
||||
for name, module in self.named_children():
|
||||
fn_recursive_attn_processor(name, module, processor)
|
||||
|
||||
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_default_attn_processor
|
||||
def set_default_attn_processor(self):
|
||||
"""
|
||||
Disables custom attention processors and sets the default attention implementation.
|
||||
"""
|
||||
if all(proc.__class__ in ADDED_KV_ATTENTION_PROCESSORS for proc in self.attn_processors.values()):
|
||||
processor = AttnAddedKVProcessor()
|
||||
elif all(proc.__class__ in CROSS_ATTENTION_PROCESSORS for proc in self.attn_processors.values()):
|
||||
processor = AttnProcessor()
|
||||
else:
|
||||
raise ValueError(
|
||||
f"Cannot call `set_default_attn_processor` when attention processors are of type {next(iter(self.attn_processors.values()))}"
|
||||
)
|
||||
|
||||
self.set_attn_processor(processor)
|
||||
|
||||
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attention_slice
|
||||
def set_attention_slice(self, slice_size: Union[str, int, List[int]]) -> None:
|
||||
r"""
|
||||
Enable sliced attention computation.
|
||||
|
||||
When this option is enabled, the attention module splits the input tensor in slices to compute attention in
|
||||
several steps. This is useful for saving some memory in exchange for a small decrease in speed.
|
||||
|
||||
Args:
|
||||
slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`):
|
||||
When `"auto"`, input to the attention heads is halved, so attention is computed in two steps. If
|
||||
`"max"`, maximum amount of memory is saved by running only one slice at a time. If a number is
|
||||
provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim`
|
||||
must be a multiple of `slice_size`.
|
||||
"""
|
||||
sliceable_head_dims = []
|
||||
|
||||
def fn_recursive_retrieve_sliceable_dims(module: torch.nn.Module):
|
||||
if hasattr(module, "set_attention_slice"):
|
||||
sliceable_head_dims.append(module.sliceable_head_dim)
|
||||
|
||||
for child in module.children():
|
||||
fn_recursive_retrieve_sliceable_dims(child)
|
||||
|
||||
# retrieve number of attention layers
|
||||
for module in self.children():
|
||||
fn_recursive_retrieve_sliceable_dims(module)
|
||||
|
||||
num_sliceable_layers = len(sliceable_head_dims)
|
||||
|
||||
if slice_size == "auto":
|
||||
# half the attention head size is usually a good trade-off between
|
||||
# speed and memory
|
||||
slice_size = [dim // 2 for dim in sliceable_head_dims]
|
||||
elif slice_size == "max":
|
||||
# make smallest slice possible
|
||||
slice_size = num_sliceable_layers * [1]
|
||||
|
||||
slice_size = num_sliceable_layers * [slice_size] if not isinstance(slice_size, list) else slice_size
|
||||
|
||||
if len(slice_size) != len(sliceable_head_dims):
|
||||
raise ValueError(
|
||||
f"You have provided {len(slice_size)}, but {self.config} has {len(sliceable_head_dims)} different"
|
||||
f" attention layers. Make sure to match `len(slice_size)` to be {len(sliceable_head_dims)}."
|
||||
)
|
||||
|
||||
for i in range(len(slice_size)):
|
||||
size = slice_size[i]
|
||||
dim = sliceable_head_dims[i]
|
||||
if size is not None and size > dim:
|
||||
raise ValueError(f"size {size} has to be smaller or equal to {dim}.")
|
||||
|
||||
# Recursively walk through all the children.
|
||||
# Any children which exposes the set_attention_slice method
|
||||
# gets the message
|
||||
def fn_recursive_set_attention_slice(module: torch.nn.Module, slice_size: List[int]):
|
||||
if hasattr(module, "set_attention_slice"):
|
||||
module.set_attention_slice(slice_size.pop())
|
||||
|
||||
for child in module.children():
|
||||
fn_recursive_set_attention_slice(child, slice_size)
|
||||
|
||||
reversed_slice_size = list(reversed(slice_size))
|
||||
for module in self.children():
|
||||
fn_recursive_set_attention_slice(module, reversed_slice_size)
|
||||
|
||||
def _set_gradient_checkpointing(self, module, value: bool = False) -> None:
|
||||
if isinstance(module, (CrossAttnDownBlock2D, DownBlock2D)):
|
||||
module.gradient_checkpointing = value
|
||||
|
||||
def forward(
|
||||
self,
|
||||
sample: torch.FloatTensor,
|
||||
timestep: Union[torch.Tensor, float, int],
|
||||
encoder_hidden_states: torch.Tensor,
|
||||
brushnet_cond: torch.FloatTensor,
|
||||
conditioning_scale: float = 1.0,
|
||||
class_labels: Optional[torch.Tensor] = None,
|
||||
timestep_cond: Optional[torch.Tensor] = None,
|
||||
attention_mask: Optional[torch.Tensor] = None,
|
||||
added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
|
||||
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
||||
guess_mode: bool = False,
|
||||
return_dict: bool = True,
|
||||
) -> Union[BrushNetOutput, Tuple[Tuple[torch.FloatTensor, ...], torch.FloatTensor]]:
|
||||
"""
|
||||
The [`BrushNetCAModel`] forward method.
|
||||
|
||||
Args:
|
||||
sample (`torch.FloatTensor`):
|
||||
The noisy input tensor.
|
||||
timestep (`Union[torch.Tensor, float, int]`):
|
||||
The number of timesteps to denoise an input.
|
||||
encoder_hidden_states (`torch.Tensor`):
|
||||
The encoder hidden states.
|
||||
brushnet_cond (`torch.FloatTensor`):
|
||||
The conditional input tensor of shape `(batch_size, sequence_length, hidden_size)`.
|
||||
conditioning_scale (`float`, defaults to `1.0`):
|
||||
The scale factor for BrushNet outputs.
|
||||
class_labels (`torch.Tensor`, *optional*, defaults to `None`):
|
||||
Optional class labels for conditioning. Their embeddings will be summed with the timestep embeddings.
|
||||
timestep_cond (`torch.Tensor`, *optional*, defaults to `None`):
|
||||
Additional conditional embeddings for timestep. If provided, the embeddings will be summed with the
|
||||
timestep_embedding passed through the `self.time_embedding` layer to obtain the final timestep
|
||||
embeddings.
|
||||
attention_mask (`torch.Tensor`, *optional*, defaults to `None`):
|
||||
An attention mask of shape `(batch, key_tokens)` is applied to `encoder_hidden_states`. If `1` the mask
|
||||
is kept, otherwise if `0` it is discarded. Mask will be converted into a bias, which adds large
|
||||
negative values to the attention scores corresponding to "discard" tokens.
|
||||
added_cond_kwargs (`dict`):
|
||||
Additional conditions for the Stable Diffusion XL UNet.
|
||||
cross_attention_kwargs (`dict[str]`, *optional*, defaults to `None`):
|
||||
A kwargs dictionary that if specified is passed along to the `AttnProcessor`.
|
||||
guess_mode (`bool`, defaults to `False`):
|
||||
In this mode, the BrushNet encoder tries its best to recognize the input content of the input even if
|
||||
you remove all prompts. A `guidance_scale` between 3.0 and 5.0 is recommended.
|
||||
return_dict (`bool`, defaults to `True`):
|
||||
Whether or not to return a [`~models.brushnet.BrushNetOutput`] instead of a plain tuple.
|
||||
|
||||
Returns:
|
||||
[`~models.brushnet.BrushNetOutput`] **or** `tuple`:
|
||||
If `return_dict` is `True`, a [`~models.brushnet.BrushNetOutput`] is returned, otherwise a tuple is
|
||||
returned where the first element is the sample tensor.
|
||||
"""
|
||||
# check channel order
|
||||
channel_order = self.config.brushnet_conditioning_channel_order
|
||||
|
||||
if channel_order == "rgb":
|
||||
# in rgb order by default
|
||||
...
|
||||
elif channel_order == "bgr":
|
||||
brushnet_cond = torch.flip(brushnet_cond, dims=[1])
|
||||
else:
|
||||
raise ValueError(f"unknown `brushnet_conditioning_channel_order`: {channel_order}")
|
||||
|
||||
# prepare attention_mask
|
||||
if attention_mask is not None:
|
||||
attention_mask = (1 - attention_mask.to(sample.dtype)) * -10000.0
|
||||
attention_mask = attention_mask.unsqueeze(1)
|
||||
|
||||
# 1. time
|
||||
timesteps = timestep
|
||||
if not torch.is_tensor(timesteps):
|
||||
# TODO: this requires sync between CPU and GPU. So try to pass timesteps as tensors if you can
|
||||
# This would be a good case for the `match` statement (Python 3.10+)
|
||||
is_mps = sample.device.type == "mps"
|
||||
if isinstance(timestep, float):
|
||||
dtype = torch.float32 if is_mps else torch.float64
|
||||
else:
|
||||
dtype = torch.int32 if is_mps else torch.int64
|
||||
timesteps = torch.tensor([timesteps], dtype=dtype, device=sample.device)
|
||||
elif len(timesteps.shape) == 0:
|
||||
timesteps = timesteps[None].to(sample.device)
|
||||
|
||||
# broadcast to batch dimension in a way that's compatible with ONNX/Core ML
|
||||
timesteps = timesteps.expand(sample.shape[0])
|
||||
|
||||
t_emb = self.time_proj(timesteps)
|
||||
|
||||
# timesteps does not contain any weights and will always return f32 tensors
|
||||
# but time_embedding might actually be running in fp16. so we need to cast here.
|
||||
# there might be better ways to encapsulate this.
|
||||
t_emb = t_emb.to(dtype=sample.dtype)
|
||||
|
||||
emb = self.time_embedding(t_emb, timestep_cond)
|
||||
aug_emb = None
|
||||
|
||||
if self.class_embedding is not None:
|
||||
if class_labels is None:
|
||||
raise ValueError("class_labels should be provided when num_class_embeds > 0")
|
||||
|
||||
if self.config.class_embed_type == "timestep":
|
||||
class_labels = self.time_proj(class_labels)
|
||||
|
||||
class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
|
||||
emb = emb + class_emb
|
||||
|
||||
if self.config.addition_embed_type is not None:
|
||||
if self.config.addition_embed_type == "text":
|
||||
aug_emb = self.add_embedding(encoder_hidden_states)
|
||||
|
||||
elif self.config.addition_embed_type == "text_time":
|
||||
if "text_embeds" not in added_cond_kwargs:
|
||||
raise ValueError(
|
||||
f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `text_embeds` to be passed in `added_cond_kwargs`"
|
||||
)
|
||||
text_embeds = added_cond_kwargs.get("text_embeds")
|
||||
if "time_ids" not in added_cond_kwargs:
|
||||
raise ValueError(
|
||||
f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `time_ids` to be passed in `added_cond_kwargs`"
|
||||
)
|
||||
time_ids = added_cond_kwargs.get("time_ids")
|
||||
time_embeds = self.add_time_proj(time_ids.flatten())
|
||||
time_embeds = time_embeds.reshape((text_embeds.shape[0], -1))
|
||||
|
||||
add_embeds = torch.concat([text_embeds, time_embeds], dim=-1)
|
||||
add_embeds = add_embeds.to(emb.dtype)
|
||||
aug_emb = self.add_embedding(add_embeds)
|
||||
|
||||
emb = emb + aug_emb if aug_emb is not None else emb
|
||||
|
||||
# 2. pre-process
|
||||
brushnet_cond=torch.concat([sample,brushnet_cond],1)
|
||||
sample = self.conv_in_condition(brushnet_cond)
|
||||
|
||||
|
||||
# 3. down
|
||||
down_block_res_samples = (sample,)
|
||||
for downsample_block in self.down_blocks:
|
||||
if hasattr(downsample_block, "has_cross_attention") and downsample_block.has_cross_attention:
|
||||
sample, res_samples = downsample_block(
|
||||
hidden_states=sample,
|
||||
temb=emb,
|
||||
encoder_hidden_states=encoder_hidden_states,
|
||||
attention_mask=attention_mask,
|
||||
cross_attention_kwargs=cross_attention_kwargs,
|
||||
)
|
||||
else:
|
||||
sample, res_samples = downsample_block(hidden_states=sample, temb=emb)
|
||||
|
||||
down_block_res_samples += res_samples
|
||||
|
||||
# 4. PaintingNet down blocks
|
||||
brushnet_down_block_res_samples = ()
|
||||
for down_block_res_sample, brushnet_down_block in zip(down_block_res_samples, self.brushnet_down_blocks):
|
||||
down_block_res_sample = brushnet_down_block(down_block_res_sample)
|
||||
brushnet_down_block_res_samples = brushnet_down_block_res_samples + (down_block_res_sample,)
|
||||
|
||||
|
||||
# 5. mid
|
||||
if self.mid_block is not None:
|
||||
if hasattr(self.mid_block, "has_cross_attention") and self.mid_block.has_cross_attention:
|
||||
sample = self.mid_block(
|
||||
sample,
|
||||
emb,
|
||||
encoder_hidden_states=encoder_hidden_states,
|
||||
attention_mask=attention_mask,
|
||||
cross_attention_kwargs=cross_attention_kwargs,
|
||||
)
|
||||
else:
|
||||
sample = self.mid_block(sample, emb)
|
||||
|
||||
# 6. BrushNet mid blocks
|
||||
brushnet_mid_block_res_sample = self.brushnet_mid_block(sample)
|
||||
|
||||
|
||||
# 7. up
|
||||
up_block_res_samples = ()
|
||||
for i, upsample_block in enumerate(self.up_blocks):
|
||||
is_final_block = i == len(self.up_blocks) - 1
|
||||
|
||||
res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
|
||||
down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]
|
||||
|
||||
# if we have not reached the final block and need to forward the
|
||||
# upsample size, we do it here
|
||||
if not is_final_block:
|
||||
upsample_size = down_block_res_samples[-1].shape[2:]
|
||||
|
||||
if hasattr(upsample_block, "has_cross_attention") and upsample_block.has_cross_attention:
|
||||
sample, up_res_samples = upsample_block(
|
||||
hidden_states=sample,
|
||||
temb=emb,
|
||||
res_hidden_states_tuple=res_samples,
|
||||
encoder_hidden_states=encoder_hidden_states,
|
||||
cross_attention_kwargs=cross_attention_kwargs,
|
||||
upsample_size=upsample_size,
|
||||
attention_mask=attention_mask,
|
||||
return_res_samples=True
|
||||
)
|
||||
else:
|
||||
sample, up_res_samples = upsample_block(
|
||||
hidden_states=sample,
|
||||
temb=emb,
|
||||
res_hidden_states_tuple=res_samples,
|
||||
upsample_size=upsample_size,
|
||||
return_res_samples=True
|
||||
)
|
||||
|
||||
up_block_res_samples += up_res_samples
|
||||
|
||||
# 8. BrushNet up blocks
|
||||
brushnet_up_block_res_samples = ()
|
||||
for up_block_res_sample, brushnet_up_block in zip(up_block_res_samples, self.brushnet_up_blocks):
|
||||
up_block_res_sample = brushnet_up_block(up_block_res_sample)
|
||||
brushnet_up_block_res_samples = brushnet_up_block_res_samples + (up_block_res_sample,)
|
||||
|
||||
# 6. scaling
|
||||
if guess_mode and not self.config.global_pool_conditions:
|
||||
scales = torch.logspace(-1, 0, len(brushnet_down_block_res_samples) + 1 + len(brushnet_up_block_res_samples), device=sample.device) # 0.1 to 1.0
|
||||
scales = scales * conditioning_scale
|
||||
|
||||
brushnet_down_block_res_samples = [sample * scale for sample, scale in zip(brushnet_down_block_res_samples, scales[:len(brushnet_down_block_res_samples)])]
|
||||
brushnet_mid_block_res_sample = brushnet_mid_block_res_sample * scales[len(brushnet_down_block_res_samples)]
|
||||
brushnet_up_block_res_samples = [sample * scale for sample, scale in zip(brushnet_up_block_res_samples, scales[len(brushnet_down_block_res_samples)+1:])]
|
||||
else:
|
||||
brushnet_down_block_res_samples = [sample * conditioning_scale for sample in brushnet_down_block_res_samples]
|
||||
brushnet_mid_block_res_sample = brushnet_mid_block_res_sample * conditioning_scale
|
||||
brushnet_up_block_res_samples = [sample * conditioning_scale for sample in brushnet_up_block_res_samples]
|
||||
|
||||
|
||||
if self.config.global_pool_conditions:
|
||||
brushnet_down_block_res_samples = [
|
||||
torch.mean(sample, dim=(2, 3), keepdim=True) for sample in brushnet_down_block_res_samples
|
||||
]
|
||||
brushnet_mid_block_res_sample = torch.mean(brushnet_mid_block_res_sample, dim=(2, 3), keepdim=True)
|
||||
brushnet_up_block_res_samples = [
|
||||
torch.mean(sample, dim=(2, 3), keepdim=True) for sample in brushnet_up_block_res_samples
|
||||
]
|
||||
|
||||
if not return_dict:
|
||||
return (brushnet_down_block_res_samples, brushnet_mid_block_res_sample, brushnet_up_block_res_samples)
|
||||
|
||||
return BrushNetOutput(
|
||||
down_block_res_samples=brushnet_down_block_res_samples,
|
||||
mid_block_res_sample=brushnet_mid_block_res_sample,
|
||||
up_block_res_samples=brushnet_up_block_res_samples
|
||||
)
|
||||
|
||||
|
||||
def zero_module(module):
|
||||
for p in module.parameters():
|
||||
nn.init.zeros_(p)
|
||||
return module
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
+158
-38
@@ -1,17 +1,40 @@
|
||||
from typing import Tuple
|
||||
import torch
|
||||
from diffusers import StableDiffusionBrushNetPipeline, BrushNetModel, UniPCMultistepScheduler
|
||||
import os
|
||||
import numpy as np
|
||||
import cv2
|
||||
import numpy as np
|
||||
from PIL import Image
|
||||
|
||||
import os
|
||||
import yaml
|
||||
import folder_paths
|
||||
|
||||
import importlib
|
||||
from contextlib import nullcontext
|
||||
is_accelerate_available = importlib.util.find_spec("accelerate") is not None
|
||||
if is_accelerate_available:
|
||||
from accelerate import init_empty_weights, load_checkpoint_and_dispatch
|
||||
|
||||
from comfy.model_management import load_models_gpu
|
||||
|
||||
from diffusers.loaders.single_file_utils import create_unet_diffusers_config, convert_ldm_unet_checkpoint
|
||||
from diffusers.loaders.single_file_utils import create_scheduler_from_ldm, create_diffusers_vae_model_from_ldm
|
||||
from diffusers.models.modeling_utils import load_model_dict_into_meta
|
||||
|
||||
from diffusers import UniPCMultistepScheduler
|
||||
|
||||
from .brushnet.brushnet import BrushNetModel
|
||||
from .brushnet.pipeline_brushnet import StableDiffusionBrushNetPipeline
|
||||
from .brushnet.unet_2d_condition import UNet2DConditionModel
|
||||
|
||||
from typing import Tuple
|
||||
|
||||
import warnings
|
||||
warnings.filterwarnings("ignore", category=UserWarning, module="torch")
|
||||
warnings.filterwarnings("ignore", category=UserWarning, module="safetensors")
|
||||
|
||||
|
||||
current_directory = os.path.dirname(os.path.abspath(__file__))
|
||||
original_config_file = os.path.join(current_directory, 'brushnet', 'v1-inference.yaml')
|
||||
brushnet_config_file = os.path.join(current_directory, 'brushnet', 'brushnet.json')
|
||||
torch_dtype = torch.float16
|
||||
|
||||
class BrushNetLoader:
|
||||
|
||||
@@ -19,7 +42,7 @@ class BrushNetLoader:
|
||||
def INPUT_TYPES(s):
|
||||
return {"required":
|
||||
{
|
||||
"brushnet": (inpaint_folders(), ),
|
||||
"brushnet": (inpaint_safetensors(), ),
|
||||
},
|
||||
}
|
||||
|
||||
@@ -30,26 +53,48 @@ class BrushNetLoader:
|
||||
FUNCTION = "brushnet_loading"
|
||||
|
||||
def brushnet_loading(self, brushnet):
|
||||
brushnet_path = os.path.join(folder_paths.models_dir, "inpaint", brushnet)
|
||||
print("Brushnet... ", end="")
|
||||
brush_net = BrushNetModel.from_pretrained(brushnet_path, torch_dtype=torch.float16)
|
||||
print("loaded")
|
||||
return (brush_net,)
|
||||
brushnet_file = os.path.join(folder_paths.models_dir, "inpaint", brushnet)
|
||||
|
||||
with init_empty_weights():
|
||||
brushnet_config = BrushNetModel.load_config(brushnet_config_file)
|
||||
brushnet_model = BrushNetModel.from_config(brushnet_config)
|
||||
|
||||
brushnet_model = load_checkpoint_and_dispatch(
|
||||
brushnet_model,
|
||||
brushnet_file,
|
||||
device_map="auto",
|
||||
max_memory=None,
|
||||
offload_folder=None,
|
||||
offload_state_dict=False,
|
||||
dtype=torch_dtype,
|
||||
force_hooks=False,
|
||||
)
|
||||
|
||||
print("BrushNet model is loaded")
|
||||
|
||||
return (brushnet_model,)
|
||||
|
||||
|
||||
def inpaint_folders():
|
||||
def inpaint_safetensors():
|
||||
inpaint_path = os.path.join(folder_paths.models_dir, 'inpaint')
|
||||
abs_list = [x[0] for x in os.walk(inpaint_path)]
|
||||
|
||||
folders = []
|
||||
for x in abs_list[1:]:
|
||||
brushnet_path = os.path.join(inpaint_path, 'brushnet')
|
||||
abs_list = []
|
||||
for x in os.walk(brushnet_path):
|
||||
for name in x[2]:
|
||||
if 'safetensors' in name:
|
||||
abs_list.append(os.path.join(x[0], name))
|
||||
names = []
|
||||
for x in abs_list:
|
||||
remain = x
|
||||
y = ''
|
||||
while remain != inpaint_path:
|
||||
remain, folder = os.path.split(remain)
|
||||
y = os.path.join(folder, y)
|
||||
folders.append(y)
|
||||
return folders
|
||||
if len(y) > 0:
|
||||
y = os.path.join(folder, y)
|
||||
else:
|
||||
y = folder
|
||||
names.append(y)
|
||||
return names
|
||||
|
||||
|
||||
class BrushNetPipeline:
|
||||
@@ -58,7 +103,9 @@ class BrushNetPipeline:
|
||||
def INPUT_TYPES(s):
|
||||
return {"required":
|
||||
{
|
||||
"model": (folder_paths.get_filename_list("checkpoints"), ),
|
||||
"model": ("MODEL",),
|
||||
"clip": ("CLIP", ),
|
||||
"vae": ("VAE", ),
|
||||
"brushnet": ("BRMODEL", ),
|
||||
},
|
||||
}
|
||||
@@ -69,22 +116,55 @@ class BrushNetPipeline:
|
||||
|
||||
FUNCTION = "pipeline_loading"
|
||||
|
||||
def pipeline_loading(self, model, brushnet):
|
||||
model_path = folder_paths.get_full_path("checkpoints", model)
|
||||
print("Model... ", end="")
|
||||
pipe = StableDiffusionBrushNetPipeline.from_single_file(
|
||||
model_path,
|
||||
brushnet=brushnet,
|
||||
torch_dtype=torch.float16,
|
||||
low_cpu_mem_usage=False,
|
||||
original_config_file=os.path.join(os.path.dirname(__file__), "v1-inference.yaml"),
|
||||
def pipeline_loading(self, model, clip, vae, brushnet):
|
||||
load_models_gpu([model, clip.load_model()])
|
||||
checkpoint = model.model.state_dict_for_saving(clip.get_sd(), vae.get_sd(), None)
|
||||
|
||||
with open(original_config_file, "r") as fp:
|
||||
original_config_data = fp.read()
|
||||
original_config = yaml.safe_load(original_config_data)
|
||||
unet = create_unet(original_config, checkpoint, torch_dtype)
|
||||
|
||||
vae = create_diffusers_vae_model_from_ldm(
|
||||
'StableDiffusionBrushNetPipeline',
|
||||
original_config,
|
||||
checkpoint,
|
||||
image_size=None,
|
||||
scaling_factor=None,
|
||||
torch_dtype=None,
|
||||
model_type=None,
|
||||
)
|
||||
pipe.to("cuda")
|
||||
|
||||
scheduler_type = "ddim"
|
||||
prediction_type = None
|
||||
scheduler_components = create_scheduler_from_ldm(
|
||||
'StableDiffusionBrushNetPipeline',
|
||||
original_config,
|
||||
checkpoint,
|
||||
scheduler_type=scheduler_type,
|
||||
prediction_type=prediction_type,
|
||||
model_type=None,
|
||||
)
|
||||
|
||||
pipe = StableDiffusionBrushNetPipeline(
|
||||
unet=unet['unet'],
|
||||
vae=vae['vae'],
|
||||
text_encoder=None,
|
||||
tokenizer=None,
|
||||
scheduler=scheduler_components['scheduler'],
|
||||
brushnet=brushnet,
|
||||
requires_safety_checker=False,
|
||||
safety_checker=None,
|
||||
feature_extractor=None
|
||||
)
|
||||
|
||||
pipe.to(dtype=torch_dtype)
|
||||
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
|
||||
pipe.enable_model_cpu_offload()
|
||||
print("loaded")
|
||||
return (pipe,)
|
||||
|
||||
print("BrushNet pipeline is loaded")
|
||||
|
||||
return (pipe,)
|
||||
|
||||
|
||||
class BrushNetInpaint:
|
||||
@@ -96,7 +176,8 @@ class BrushNetInpaint:
|
||||
"BRPL": ("BRPL", ),
|
||||
"image": ("IMAGE",),
|
||||
"mask": ("MASK",),
|
||||
"prompt": ("STRING", {"multiline": False, "default": ""}),
|
||||
"positive": ("CONDITIONING", ),
|
||||
"negative": ("CONDITIONING", ),
|
||||
"seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
|
||||
"steps": ("INT", {"default": 50, "min": 1, "max": 10000}),
|
||||
"scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 10.0}),
|
||||
@@ -109,10 +190,9 @@ class BrushNetInpaint:
|
||||
|
||||
FUNCTION = "brushnet_inpaint"
|
||||
|
||||
def brushnet_inpaint(self, BRPL, image: torch.Tensor, mask: torch.Tensor, prompt: str, seed: int,
|
||||
def brushnet_inpaint(self, BRPL, image: torch.Tensor, mask: torch.Tensor, positive, negative, seed: int,
|
||||
steps: int, scale: float) -> Tuple[torch.Tensor]:
|
||||
|
||||
|
||||
print("Working on image")
|
||||
|
||||
init_image = image_from_tensor(image[0])
|
||||
@@ -128,11 +208,20 @@ class BrushNetInpaint:
|
||||
|
||||
print("Inference started")
|
||||
|
||||
pos_embeds = positive
|
||||
while isinstance(pos_embeds, list):
|
||||
pos_embeds = pos_embeds[0]
|
||||
|
||||
neg_embeds = negative
|
||||
while isinstance(neg_embeds, list):
|
||||
neg_embeds = neg_embeds[0]
|
||||
|
||||
generator = torch.Generator("cuda").manual_seed(seed)
|
||||
result = BRPL(
|
||||
prompt,
|
||||
init_image,
|
||||
mask_image,
|
||||
image=init_image,
|
||||
mask=mask_image,
|
||||
prompt_embeds=pos_embeds,
|
||||
negative_prompt_embeds=neg_embeds,
|
||||
num_inference_steps=steps,
|
||||
generator=generator,
|
||||
brushnet_conditioning_scale=scale
|
||||
@@ -190,3 +279,34 @@ def numpy_to_tensor(array: np.ndarray) -> torch.Tensor:
|
||||
"""Convert a numpy array to a tensor and scale its values from 0-255 to 0-1."""
|
||||
array = array.astype(np.float32) / 255.0
|
||||
return torch.from_numpy(array)[None,]
|
||||
|
||||
def create_unet(original_config, checkpoint, torch_dtype):
|
||||
num_in_channels = 4
|
||||
image_size = 512
|
||||
|
||||
unet_config = create_unet_diffusers_config(original_config, image_size=image_size)
|
||||
unet_config["in_channels"] = num_in_channels
|
||||
diffusers_format_unet_checkpoint = convert_ldm_unet_checkpoint(checkpoint, unet_config, extract_ema=False)
|
||||
|
||||
ctx = init_empty_weights if is_accelerate_available else nullcontext
|
||||
|
||||
with ctx():
|
||||
unet = UNet2DConditionModel(**unet_config)
|
||||
|
||||
if is_accelerate_available:
|
||||
unexpected_keys = load_model_dict_into_meta(unet, diffusers_format_unet_checkpoint, dtype=torch_dtype)
|
||||
if unet._keys_to_ignore_on_load_unexpected is not None:
|
||||
for pat in unet._keys_to_ignore_on_load_unexpected:
|
||||
unexpected_keys = [k for k in unexpected_keys if re.search(pat, k) is None]
|
||||
|
||||
if len(unexpected_keys) > 0:
|
||||
print(
|
||||
f"Some weights of the model checkpoint were not used when initializing {unet.__name__}: \n {[', '.join(unexpected_keys)]}"
|
||||
)
|
||||
else:
|
||||
unet.load_state_dict(diffusers_format_unet_checkpoint)
|
||||
|
||||
if torch_dtype is not None:
|
||||
unet = unet.to(torch_dtype)
|
||||
|
||||
return {"unet": unet}
|
||||
|
||||
File diff suppressed because one or more lines are too long
Binary file not shown.
|
After Width: | Height: | Size: 1.6 MiB |
File diff suppressed because one or more lines are too long
Binary file not shown.
|
After Width: | Height: | Size: 1.7 MiB |
@@ -1 +0,0 @@
|
||||
{"last_node_id": 44, "last_link_id": 68, "nodes": [{"id": 29, "type": "GroundingDinoSAMSegment (segment anything)", "pos": [749, 640], "size": {"0": 352.79998779296875, "1": 124.0001220703125}, "flags": {}, "order": 5, "mode": 0, "inputs": [{"name": "sam_model", "type": "SAM_MODEL", "link": 55}, {"name": "grounding_dino_model", "type": "GROUNDING_DINO_MODEL", "link": 56}, {"name": "image", "type": "IMAGE", "link": 57}], "outputs": [{"name": "IMAGE", "type": "IMAGE", "links": [], "shape": 3, "slot_index": 0}, {"name": "MASK", "type": "MASK", "links": [59, 61], "shape": 3, "slot_index": 1}], "properties": {"Node name for S&R": "GroundingDinoSAMSegment (segment anything)"}, "widgets_values": ["cake", 0.3]}, {"id": 27, "type": "GroundingDinoModelLoader (segment anything)", "pos": [45, 1126], "size": {"0": 532.800048828125, "1": 104.4000244140625}, "flags": {}, "order": 2, "mode": 0, "outputs": [{"name": "GROUNDING_DINO_MODEL", "type": "GROUNDING_DINO_MODEL", "links": [56], "shape": 3, "slot_index": 0}], "properties": {"Node name for S&R": "GroundingDinoModelLoader (segment anything)"}, "widgets_values": ["GroundingDINO_SwinT_OGC (694MB)"]}, {"id": 28, "type": "SAMModelLoader (segment anything)", "pos": [254, 1014], "size": {"0": 315, "1": 58}, "flags": {}, "order": 1, "mode": 0, "outputs": [{"name": "SAM_MODEL", "type": "SAM_MODEL", "links": [55], "shape": 3, "slot_index": 0}], "properties": {"Node name for S&R": "SAMModelLoader (segment anything)"}, "widgets_values": ["sam_vit_h (2.56GB)"]}, {"id": 1, "type": "LoadImage", "pos": [101, 386], "size": [470.19439697265625, 578.6853942871094], "flags": {}, "order": 3, "mode": 0, "outputs": [{"name": "IMAGE", "type": "IMAGE", "links": [52, 57], "shape": 3, "slot_index": 0}, {"name": "MASK", "type": "MASK", "links": null, "shape": 3, "slot_index": 1}], "properties": {"Node name for S&R": "LoadImage"}, "widgets_values": ["test_image.jpg", "image"]}, {"id": 24, "type": "BrushNetLoader", "pos": [222, 260], "size": {"0": 354.79998779296875, "1": 61.60003662109375}, "flags": {}, "order": 0, "mode": 0, "outputs": [{"name": "BrushNet", "type": "BRMODEL", "links": [67], "shape": 3, "slot_index": 0}], "properties": {"Node name for S&R": "BrushNetLoader"}, "widgets_values": ["brushnet/random_mask_brushnet_ckpt"]}, {"id": 44, "type": "BrushNetPipeline", "pos": [704, 263], "size": [371.60003662109375, 103.20001220703125], "flags": {}, "order": 4, "mode": 0, "inputs": [{"name": "brushnet", "type": "BRMODEL", "link": 67}], "outputs": [{"name": "BRPL", "type": "BRPL", "links": [66], "shape": 3, "slot_index": 0}], "properties": {"Node name for S&R": "BrushNetPipeline"}, "widgets_values": ["realisticVisionV60B1_v51VAE.safetensors"]}, {"id": 26, "type": "BrushNetInpaint", "pos": [1254, 411], "size": {"0": 315, "1": 194}, "flags": {}, "order": 7, "mode": 0, "inputs": [{"name": "BRPL", "type": "BRPL", "link": 66}, {"name": "image", "type": "IMAGE", "link": 52}, {"name": "mask", "type": "MASK", "link": 61}], "outputs": [{"name": "Image", "type": "IMAGE", "links": [53], "shape": 3, "slot_index": 0}], "properties": {"Node name for S&R": "BrushNetInpaint"}, "widgets_values": ["burger", 0, "fixed", 30, 1]}, {"id": 12, "type": "PreviewImage", "pos": [1634, 411], "size": [523.5944213867188, 547.4853210449219], "flags": {}, "order": 9, "mode": 0, "inputs": [{"name": "images", "type": "IMAGE", "link": 53}], "properties": {"Node name for S&R": "PreviewImage"}}, {"id": 32, "type": "PreviewImage", "pos": [1386, 711], "size": {"0": 210, "1": 246}, "flags": {}, "order": 8, "mode": 0, "inputs": [{"name": "images", "type": "IMAGE", "link": 60}], "properties": {"Node name for S&R": "PreviewImage"}}, {"id": 31, "type": "MaskToImage", "pos": [1140, 711], "size": {"0": 210, "1": 26}, "flags": {}, "order": 6, "mode": 0, "inputs": [{"name": "mask", "type": "MASK", "link": 59}], "outputs": [{"name": "IMAGE", "type": "IMAGE", "links": [60], "shape": 3, "slot_index": 0}], "properties": {"Node name for S&R": "MaskToImage"}}], "links": [[52, 1, 0, 26, 1, "IMAGE"], [53, 26, 0, 12, 0, "IMAGE"], [55, 28, 0, 29, 0, "SAM_MODEL"], [56, 27, 0, 29, 1, "GROUNDING_DINO_MODEL"], [57, 1, 0, 29, 2, "IMAGE"], [59, 29, 1, 31, 0, "MASK"], [60, 31, 0, 32, 0, "IMAGE"], [61, 29, 1, 26, 2, "MASK"], [66, 44, 0, 26, 0, "BRPL"], [67, 24, 0, 44, 0, "BRMODEL"]], "groups": [], "config": {}, "extra": {}, "version": 0.4}
|
||||
Binary file not shown.
|
Before Width: | Height: | Size: 1.5 MiB |
Binary file not shown.
|
Before Width: | Height: | Size: 34 KiB |
Binary file not shown.
|
Before Width: | Height: | Size: 11 KiB |
+3
-2
@@ -1,6 +1,7 @@
|
||||
torch
|
||||
diffusers
|
||||
accelerate
|
||||
numpy
|
||||
opencv-python
|
||||
Pillow
|
||||
accelerate
|
||||
git+https://github.com/nullquant/diffusers.git
|
||||
yaml
|
||||
Reference in New Issue
Block a user