# Copyright 2023 The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. 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 einops import rearrange, repeat from diffusers.configuration_utils import ConfigMixin, register_to_config from diffusers.utils import BaseOutput, logging from diffusers.models.attention_processor import AttentionProcessor, AttnProcessor # from diffusers.models.transformer_temporal import TransformerTemporalModel from diffusers.models.embeddings import TimestepEmbedding, Timesteps from diffusers.models.modeling_utils import ModelMixin from t2v_enhanced.model.diffusers_conditional.models.controlnet.unet_3d_blocks import ( CrossAttnDownBlock3D, CrossAttnUpBlock3D, DownBlock3D, UNetMidBlock3DCrossAttn, UpBlock3D, get_down_block, get_up_block, transformer_g_c ) # from diffusers.models.unet_3d_condition import UNet3DConditionModel from t2v_enhanced.model.diffusers_conditional.models.controlnet.unet_3d_condition import UNet3DConditionModel from t2v_enhanced.model.diffusers_conditional.models.controlnet.transformer_temporal import TransformerTemporalModel from t2v_enhanced.model.layers.conv_channel_extension import Conv2D_SubChannels logger = logging.get_logger(__name__) # pylint: disable=invalid-name @dataclass class ControlNetOutput(BaseOutput): down_block_res_samples: Tuple[torch.Tensor] mid_block_res_sample: torch.Tensor class Merger(nn.Module): def __init__(self, n_frames_condition: int = 8, n_frames_sample: int = 16, merge_mode: str = "addition", input_channels=0, frame_expansion="last_frame") -> None: super().__init__() self.merge_mode = merge_mode self.n_frames_condition = n_frames_condition self.n_frames_sample = n_frames_sample self.frame_expansion = frame_expansion if merge_mode.startswith("attention"): self.attention = ConditionalModel(input_channels=input_channels, conditional_model=merge_mode.split("attention_")[1]) def forward(self, x, condition_signal): x = rearrange(x, "(B F) C H W -> B F C H W", F=self.n_frames_sample) condition_signal = rearrange( condition_signal, "(B F) C H W -> B F C H W", B=x.shape[0]) if x.shape[1] - condition_signal.shape[1] > 0: if self.frame_expansion == "last_frame": fillup_latent = repeat( condition_signal[:, -1], "B C H W -> B F C H W", F=x.shape[1] - condition_signal.shape[1]) elif self.frame_expansion == "zero": fillup_latent = torch.zeros( (x.shape[0], self.n_frames_sample-self.n_frames_condition, *x.shape[2:]), device=x.device, dtype=x.dtype) if self.frame_expansion != "none": condition_signal = torch.cat( [condition_signal, fillup_latent], dim=1) if self.merge_mode == "addition": out = x + condition_signal elif self.merge_mode.startswith("attention"): out = self.attention(x, condition_signal) out = rearrange(out, "B F C H W -> (B F) C H W") return out class ZeroConv(nn.Module): def __init__(self, channels: int, mode: str = "2d", num_frames: int = 8, zero_init=True): super().__init__() mode_parts = mode.split("_") if len(mode_parts) > 1 and mode_parts[1] == "noinit": zero_init = False if mode.startswith("2d"): model = nn.Conv2d( channels, channels, kernel_size=1) model = zero_module(model, reset=zero_init) elif mode.startswith("3d"): model = ZeroConv3D(num_frames=num_frames, channels=channels, zero_init=zero_init) elif mode == "Identity": model = nn.Identity() self.model = model def forward(self, x): return self.model(x) class ControlNetConditioningEmbedding(nn.Module): """ Quoting from https://arxiv.org/abs/2302.05543: "Stable Diffusion uses a pre-processing method similar to VQ-GAN [11] to convert the entire dataset of 512 × 512 images into smaller 64 × 64 “latent images” for stabilized training. This requires ControlNets to convert image-based conditions to 64 × 64 feature space to match the convolution size. We use a tiny network E(·) of four convolution layers with 4 × 4 kernels and 2 × 2 strides (activated by ReLU, channels are 16, 32, 64, 128, initialized with Gaussian weights, trained jointly with the full model) to encode image-space conditions ... into feature maps ..." """ # TODO why not GAUSSIAN used? # TODO why not 4x4 kernel? # TODO why not 2 x2 stride? def __init__( self, conditioning_embedding_channels: int, conditioning_channels: int = 3, block_out_channels: Tuple[int] = (16, 32, 96, 256), downsample: bool = True, final_3d_conv: bool = False, num_frame_conditioning: int = 8, num_frames: int = 16, zero_init: bool = True, use_controlnet_mask: bool = False, use_normalization: bool = False, ): super().__init__() self.num_frame_conditioning = num_frame_conditioning self.num_frames = num_frames self.final_3d_conv = final_3d_conv self.conv_in = nn.Conv2d( conditioning_channels, block_out_channels[0], kernel_size=3, padding=1) if final_3d_conv: print("USING 3D CONV in ControlNET") self.blocks = nn.ModuleList([]) if use_normalization: self.norms = nn.ModuleList([]) self.use_normalization = use_normalization stride = 2 if downsample else 1 if use_normalization: res = 256 # HARD-CODED Resolution! for i in range(len(block_out_channels) - 1): channel_in = block_out_channels[i] channel_out = block_out_channels[i + 1] self.blocks.append( nn.Conv2d(channel_in, channel_in, kernel_size=3, padding=1)) if use_normalization: self.norms.append(nn.LayerNorm((channel_in, res, res))) self.blocks.append( nn.Conv2d(channel_in, channel_out, kernel_size=3, padding=1, stride=stride)) if use_normalization: res = res // 2 self.norms.append(nn.LayerNorm((channel_out, res, res))) if not final_3d_conv: self.conv_out = zero_module( nn.Conv2d( block_out_channels[-1]+int(use_controlnet_mask), conditioning_embedding_channels, kernel_size=3, padding=1), reset=zero_init ) else: self.conv_temp = zero_module(TemporalConvLayer_Custom( num_frame_conditioning, num_frames, dropout=0.0), reset=zero_init) self.conv_out = nn.Conv2d( block_out_channels[-1]+int(use_controlnet_mask), conditioning_embedding_channels, kernel_size=3, padding=1) # self.conv_temp = zero_module(nn.Conv3d( # num_frame_conditioning, num_frames, kernel_size=3, padding=1) # ) def forward(self, conditioning, vq_gan=None, controlnet_mask=None): embedding = self.conv_in(conditioning) embedding = F.silu(embedding) if self.use_normalization: for block, norm in zip(self.blocks, self.norms): embedding = block(embedding) embedding = norm(embedding) embedding = F.silu(embedding) else: for block in self.blocks: embedding = block(embedding) embedding = F.silu(embedding) if controlnet_mask is not None: embedding = rearrange( embedding, "(B F) C H W -> F B C H W", F=self.num_frames) controlnet_mask_expanded = controlnet_mask[:, :, None, None, None] controlnet_mask_expanded = rearrange( controlnet_mask_expanded, "B F C W H -> F B C W H") masked_embedding = controlnet_mask_expanded * embedding embedding = rearrange(masked_embedding, "F B C H W -> (B F) C H W") controlnet_mask_expanded = rearrange( controlnet_mask_expanded, "F B C H W -> (B F) C H W") # controlnet_mask_expanded = repeat(controlnet_mask_expanded,"B C W H -> B (C x) W H",x=embedding.shape[1]) controlnet_mask_expanded = repeat( controlnet_mask_expanded, "B C W H -> B C (W y) H", y=embedding.shape[2]) controlnet_mask_expanded = repeat( controlnet_mask_expanded, "B C W H -> B C W (H z)", z=embedding.shape[3]) embedding = torch.cat([embedding, controlnet_mask_expanded], dim=1) embedding = self.conv_out(embedding) if self.final_3d_conv: # embedding = F.silu(embedding) embedding = rearrange( embedding, "(b f) c h w -> b f c h w", f=self.num_frame_conditioning) embedding = self.conv_temp(embedding) embedding = rearrange(embedding, "b f c h w -> (b f) c h w") return embedding class ControlNetModel(ModelMixin, ConfigMixin): _supports_gradient_checkpointing = False @register_to_config def __init__( self, in_channels: int = 4, flip_sin_to_cos: bool = True, freq_shift: int = 0, down_block_types: Tuple[str] = ( "CrossAttnDownBlock3D", "CrossAttnDownBlock3D", "CrossAttnDownBlock3D", "DownBlock3D", ), 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, attention_head_dim: Union[int, Tuple[int]] = 8, use_linear_projection: bool = False, class_embed_type: Optional[str] = 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, controlnet_conditioning_channel_order: str = "rgb", conditioning_embedding_out_channels: Optional[Tuple[int]] = ( 16, 32, 96, 256), global_pool_conditions: bool = False, downsample_controlnet_cond: bool = True, frame_expansion: str = "zero", condition_encoder: str = "", num_frames: int = 16, num_frame_conditioning: int = 8, num_tranformers: int = 1, vae=None, merging_mode: str = "addition", zero_conv_mode: str = "2d", use_controlnet_mask: bool = False, use_image_embedding: bool = False, use_image_encoder_normalization: bool = False, unet_params=None, ): super().__init__() self.gradient_checkpointing = False # Check inputs 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(attention_head_dim, int) and len(attention_head_dim) != len(down_block_types): raise ValueError( f"Must provide the same number of `attention_head_dim` as `down_block_types`. `attention_head_dim`: {attention_head_dim}. `down_block_types`: {down_block_types}." ) self.use_image_tokens = unet_params.use_image_tokens_ctrl self.image_encoder_name = type(unet_params.image_encoder).__name__ # input conv_in_kernel = 3 conv_in_padding = (conv_in_kernel - 1) // 2 '''Conv2D_SubChannels self.conv_in = nn.Conv2d( in_channels, block_out_channels[0], kernel_size=conv_in_kernel, padding=conv_in_padding ) ''' self.conv_in = Conv2D_SubChannels( in_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, ) self.transformer_in = TransformerTemporalModel( num_attention_heads=8, attention_head_dim=attention_head_dim, in_channels=block_out_channels[0], num_layers=1, ) # 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 conditioning_channels = 3 if downsample_controlnet_cond else 4 # control net conditioning embedding if condition_encoder == "temp_conv_vq": controlnet_cond_embedding = ControlNetConditioningEmbeddingVQ( conditioning_embedding_channels=block_out_channels[0], conditioning_channels=4, block_out_channels=conditioning_embedding_out_channels, downsample=False, num_frame_conditioning=num_frame_conditioning, num_frames=num_frames, num_tranformers=num_tranformers, # zero_init=not merging_mode.startswith("attention"), ) elif condition_encoder == "vq": controlnet_cond_embedding = ControlNetConditioningOptVQ(vq=vae, conditioning_embedding_channels=block_out_channels[ 0], conditioning_channels=4, block_out_channels=conditioning_embedding_out_channels, num_frame_conditioning=num_frame_conditioning, num_frames=num_frames, ) else: controlnet_cond_embedding = ControlNetConditioningEmbedding( conditioning_embedding_channels=block_out_channels[0], conditioning_channels=conditioning_channels, block_out_channels=conditioning_embedding_out_channels, downsample=downsample_controlnet_cond, final_3d_conv=condition_encoder.endswith("3DConv"), num_frame_conditioning=num_frame_conditioning, num_frames=num_frames, # zero_init=not merging_mode.startswith("attention") use_controlnet_mask=use_controlnet_mask, use_normalization=use_image_encoder_normalization, ) self.use_controlnet_mask = use_controlnet_mask self.down_blocks = nn.ModuleList([]) self.controlnet_down_blocks = nn.ModuleList([]) # conv_in self.merger = Merger(n_frames_sample=num_frames, n_frames_condition=num_frame_conditioning, merge_mode=merging_mode, input_channels=block_out_channels[0], frame_expansion=frame_expansion) 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) # down output_channel = block_out_channels[0] self.controlnet_down_blocks.append( ZeroConv(channels=output_channel, mode=zero_conv_mode, num_frames=num_frames)) 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, 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, attn_num_head_channels=attention_head_dim[i], downsample_padding=downsample_padding, dual_cross_attention=False, use_image_embedding=use_image_embedding, unet_params=unet_params, ) self.down_blocks.append(down_block) for _ in range(layers_per_block): self.controlnet_down_blocks.append( ZeroConv(channels=output_channel, mode=zero_conv_mode, num_frames=num_frames)) if not is_final_block: self.controlnet_down_blocks.append( ZeroConv(channels=output_channel, mode=zero_conv_mode, num_frames=num_frames)) # mid mid_block_channel = block_out_channels[-1] self.controlnet_mid_block = ZeroConv( channels=mid_block_channel, mode=zero_conv_mode, num_frames=num_frames) self.mid_block = UNetMidBlock3DCrossAttn( in_channels=block_out_channels[-1], temb_channels=time_embed_dim, resnet_eps=norm_eps, resnet_act_fn=act_fn, output_scale_factor=mid_block_scale_factor, cross_attention_dim=cross_attention_dim, attn_num_head_channels=attention_head_dim[-1], resnet_groups=norm_num_groups, dual_cross_attention=False, use_image_embedding=use_image_embedding, unet_params=unet_params, ) self.controlnet_cond_embedding = controlnet_cond_embedding self.num_frames = num_frames self.num_frame_conditioning = num_frame_conditioning @classmethod def from_unet( cls, unet: UNet3DConditionModel, controlnet_conditioning_channel_order: str = "rgb", conditioning_embedding_out_channels: Optional[Tuple[int]] = ( 16, 32, 96, 256), load_weights_from_unet: bool = True, downsample_controlnet_cond: bool = True, num_frames: int = 16, num_frame_conditioning: int = 8, frame_expansion: str = "zero", num_tranformers: int = 1, vae=None, zero_conv_mode: str = "2d", merging_mode: str = "addition", # [spatial,spatial_3DConv,temp_conv_vq] condition_encoder: str = "spatial_3DConv", use_controlnet_mask: bool = False, use_image_embedding: bool = False, use_image_encoder_normalization: bool = False, unet_params=None, ** kwargs, ): r""" Instantiate Controlnet class from UNet3DConditionModel. Parameters: unet (`UNet3DConditionModel`): UNet model which weights are copied to the ControlNet. Note that all configuration options are also copied where applicable. """ controlnet = cls( in_channels=unet.config.in_channels, down_block_types=unet.config.down_block_types, block_out_channels=unet.config.block_out_channels, layers_per_block=unet.config.layers_per_block, 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, attention_head_dim=unet.config.attention_head_dim, conditioning_embedding_out_channels=conditioning_embedding_out_channels, downsample_controlnet_cond=downsample_controlnet_cond, num_frame_conditioning=num_frame_conditioning, num_frames=num_frames, frame_expansion=frame_expansion, num_tranformers=num_tranformers, vae=vae, zero_conv_mode=zero_conv_mode, merging_mode=merging_mode, condition_encoder=condition_encoder, use_controlnet_mask=use_controlnet_mask, use_image_embedding=use_image_embedding, use_image_encoder_normalization=use_image_encoder_normalization, unet_params=unet_params, ) if load_weights_from_unet: controlnet.conv_in.load_state_dict(unet.conv_in.state_dict()) controlnet.time_proj.load_state_dict(unet.time_proj.state_dict()) controlnet.transformer_in.load_state_dict( unet.transformer_in.state_dict()) controlnet.time_embedding.load_state_dict( unet.time_embedding.state_dict()) if controlnet.class_embedding: controlnet.class_embedding.load_state_dict( unet.class_embedding.state_dict()) controlnet.down_blocks.load_state_dict( unet.down_blocks.state_dict(), strict=False) # can be that the controlnet model does not use image clip encoding controlnet.mid_block.load_state_dict( unet.mid_block.state_dict(), strict=False) return controlnet @property # Copied from diffusers.models.unet_3d_condition.UNet3DConditionModel.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, "set_processor"): processors[f"{name}.processor"] = module.processor 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.unet_3d_condition.UNet3DConditionModel.set_attn_processor def set_attn_processor(self, processor: Union[AttentionProcessor, Dict[str, AttentionProcessor]]): r""" Parameters: `processor (`dict` of `AttentionProcessor` or `AttentionProcessor`): The instantiated processor class or a dictionary of processor classes that will be set as the processor of **all** `Attention` layers. In case `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.unet_3d_condition.UNet3DConditionModel.set_default_attn_processor def set_default_attn_processor(self): """ Disables custom attention processors and sets the default attention implementation. """ self.set_attn_processor(AttnProcessor()) # Copied from diffusers.models.unet_3d_condition.UNet3DConditionModel.set_attention_slice def set_attention_slice(self, slice_size): r""" Enable sliced attention computation. When this option is enabled, the attention module will split the input tensor in slices, to compute attention in several steps. This is useful to save some memory in exchange for a small speed decrease. Args: slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`): When `"auto"`, halves the input to the attention heads, so attention will be computed in two steps. If `"max"`, maximum amount of memory will be 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=False): if isinstance(module, (CrossAttnDownBlock3D, DownBlock3D)): module.gradient_checkpointing = value # TODO ADD WEIGHT CONTROL def forward( self, sample: torch.FloatTensor, timestep: Union[torch.Tensor, float, int], encoder_hidden_states: torch.Tensor, controlnet_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, cross_attention_kwargs: Optional[Dict[str, Any]] = None, guess_mode: bool = False, return_dict: bool = True, weight_control: float = 1.0, weight_control_sample: float = 1.0, controlnet_mask: Optional[torch.Tensor] = None, vq_gan=None, ) -> Union[ControlNetOutput, Tuple]: # check channel order # TODO SET ATTENTION MASK And WEIGHT CONTROL as in CONTROLNET.PY ''' # 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) ''' # assert controlnet_mask is None, "Controlnet Mask not implemented yet for clean model" # 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) sample = sample[:, :, :self.num_frames] # broadcast to batch dimension in a way that's compatible with ONNX/Core ML num_frames = sample.shape[2] 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=self.dtype) emb = self.time_embedding(t_emb, timestep_cond) emb = emb.repeat_interleave(repeats=num_frames, dim=0) if not self.use_image_tokens and encoder_hidden_states.shape[1] > 77: encoder_hidden_states = encoder_hidden_states[:, :77] if encoder_hidden_states.shape[1] > 77: # assert ( # encoder_hidden_states.shape[1]-77) % num_frames == 0, f"Encoder shape {encoder_hidden_states.shape}. Num frames = {num_frames}" context_text, context_img = encoder_hidden_states[:, :77, :], encoder_hidden_states[:, 77:, :] context_text = context_text.repeat_interleave( repeats=num_frames, dim=0) if self.image_encoder_name == "FrozenOpenCLIPImageEmbedder": context_img = context_img.repeat_interleave( repeats=num_frames, dim=0) else: context_img = rearrange( context_img, 'b (t l) c -> (b t) l c', t=num_frames) encoder_hidden_states = torch.cat( [context_text, context_img], dim=1) else: encoder_hidden_states = encoder_hidden_states.repeat_interleave( repeats=num_frames, dim=0) # print(f"ctrl with tokens = {encoder_hidden_states.shape[1]}") ''' encoder_hidden_states = encoder_hidden_states.repeat_interleave( repeats=num_frames, dim=0) ''' # 2. pre-process sample = sample.permute(0, 2, 1, 3, 4).reshape( (sample.shape[0] * num_frames, -1) + sample.shape[3:]) sample = self.conv_in(sample) controlnet_cond = self.controlnet_cond_embedding( controlnet_cond, vq_gan=vq_gan, controlnet_mask=controlnet_mask) if num_frames > 1: if self.gradient_checkpointing: sample = transformer_g_c( self.transformer_in, sample, num_frames) else: sample = self.transformer_in( sample, num_frames=num_frames, attention_mask=attention_mask).sample sample = self.merger(sample * weight_control_sample, weight_control * controlnet_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, num_frames=num_frames, cross_attention_kwargs=cross_attention_kwargs, ) else: sample, res_samples = downsample_block( hidden_states=sample, temb=emb, num_frames=num_frames) down_block_res_samples += res_samples # 4. mid if self.mid_block is not None: sample = self.mid_block( sample, emb, encoder_hidden_states=encoder_hidden_states, attention_mask=attention_mask, num_frames=num_frames, cross_attention_kwargs=cross_attention_kwargs, ) # 5. Control net blocks controlnet_down_block_res_samples = () for down_block_res_sample, controlnet_block in zip(down_block_res_samples, self.controlnet_down_blocks): down_block_res_sample = controlnet_block(down_block_res_sample) controlnet_down_block_res_samples = controlnet_down_block_res_samples + \ (down_block_res_sample,) down_block_res_samples = controlnet_down_block_res_samples mid_block_res_sample = self.controlnet_mid_block(sample) # 6. scaling if guess_mode and not self.config.global_pool_conditions: # 0.1 to 1.0 scales = torch.logspace(-1, 0, len(down_block_res_samples) + 1, device=sample.device) scales = scales * conditioning_scale down_block_res_samples = [ sample * scale for sample, scale in zip(down_block_res_samples, scales)] mid_block_res_sample = mid_block_res_sample * \ scales[-1] # last one else: down_block_res_samples = [ sample * conditioning_scale for sample in down_block_res_samples] mid_block_res_sample = mid_block_res_sample * conditioning_scale if self.config.global_pool_conditions: down_block_res_samples = [ torch.mean(sample, dim=(2, 3), keepdim=True) for sample in down_block_res_samples ] mid_block_res_sample = torch.mean( mid_block_res_sample, dim=(2, 3), keepdim=True) if not return_dict: return (down_block_res_samples, mid_block_res_sample) return ControlNetOutput( down_block_res_samples=down_block_res_samples, mid_block_res_sample=mid_block_res_sample ) def zero_module(module, reset=True): if reset: for p in module.parameters(): nn.init.zeros_(p) return module