model_lib/ControlNet/cldm/cldm.py

from re import I
import torch
import torch as th
import torch.nn as nn
from model_lib.ControlNet.ldm.modules.diffusionmodules.util import (
    conv_nd,
    linear,
    zero_module,
    timestep_embedding,
)

from model_lib.ControlNet.ldm.modules.attention import SpatialTransformer
from model_lib.ControlNet.ldm.modules.diffusionmodules.openaimodel import TimestepEmbedSequential, ResBlock, Downsample, AttentionBlock,Upsample, UNetModel_Temporal
from model_lib.ControlNet.ldm.models.diffusion.ddpm import LatentDiffusionReferenceOnly
from model_lib.ControlNet.ldm.util import exists, instantiate_from_config

## TODO: here UNet
class ControlledUnetModelAttn_Temporal_Pose_Local(UNetModel_Temporal):
    def forward(self, x, timesteps=None, context=None, control=None, pose_control=None,local_pose_control=None,only_mid_control=False, attention_mode=None,uc=False, **kwargs):
        hs = []
        bank_attn = control
        attn_index = 0

        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
        emb = self.time_embed(t_emb)
        h = x.type(self.dtype)
        num_input_motion_module = 0
        if uc:
            for i, module in enumerate(self.input_blocks):
                if i in [1,2,4,5,7,8,10,11]:
                    motion_module = self.input_blocks_motion_module[num_input_motion_module]
                    h = module(h, emb, context,uc=uc) # Attn here
                    h = motion_module(h, emb, context)
                    num_input_motion_module += 1
                else:
                    h = module(h, emb, context,uc=uc) # Attn here
                hs.append(h)

            h = self.middle_block(h, emb, context,uc=uc) # Attn here

            for i, module in enumerate(self.output_blocks):
                output_block_motion_module = self.output_blocks_motion_module[i]
                if only_mid_control:
                    h = torch.cat([h, hs.pop()], dim=1)
                    h = module(h, emb, context,uc=uc) 
                else:
                    h = torch.cat([h, hs.pop()], dim=1)
                    h = module(h, emb, context,uc=uc) # Attn here
                    h = output_block_motion_module(h, emb, context) 

        else:   
            num_input_motion_module = 0             
            for i, module in enumerate(self.input_blocks): 
                if i in [1,2,4,5,7,8,10,11]:
                    motion_module = self.input_blocks_motion_module[num_input_motion_module]
                    h, attn_index = module(h, emb, context, bank_attn, attention_mode, attn_index)
                    h = motion_module(h, emb, context) 
                    num_input_motion_module += 1
                else:
                    h, attn_index = module(h, emb, context, bank_attn, attention_mode, attn_index) # Attn here
                hs.append(h)
            
            h, attn_index = self.middle_block(h, emb, context, bank_attn, attention_mode, attn_index) # Attn here

            amplify_f = 1.

            if pose_control is not None:
                h += pose_control.pop() * amplify_f

            if local_pose_control is not None:
                h += local_pose_control.pop() * amplify_f

            for i, module in enumerate(self.output_blocks):
                output_block_motion_module = self.output_blocks_motion_module[i]
                if only_mid_control or (bank_attn is None):
                    h = torch.cat([h, hs.pop()], dim=1)
                    h = module(h, emb, context) 
                else:
                    if pose_control is not None and local_pose_control is not None:
                        h = torch.cat([h, hs.pop() + pose_control.pop() * amplify_f + local_pose_control.pop() * amplify_f], dim=1)
                    elif pose_control is not None:
                        h = torch.cat([h, hs.pop() + pose_control.pop() * amplify_f], dim=1)
                    elif local_pose_control is not None:
                        h = torch.cat([h, hs.pop() + local_pose_control.pop() * amplify_f], dim=1)
                    else: 
                        h = torch.cat([h, hs.pop()], dim=1)

                    h, attn_index = module(h, emb, context, bank_attn, attention_mode, attn_index) # Attn here
                    h = output_block_motion_module(h, emb, context) 

        h = h.type(x.dtype)
        return self.out(h)


## ControlNet Reference Only-Like Attention
class ControlNetReferenceOnly(nn.Module):
    def __init__(
        self,
        image_size,
        in_channels,
        model_channels,
        hint_channels,
        out_channels,
        num_res_blocks,
        attention_resolutions,
        dropout=0,
        channel_mult=(1, 2, 4, 8),
        conv_resample=True,
        dims=2,
        use_checkpoint=False,
        use_fp16=False,
        num_heads=-1,
        num_head_channels=-1,
        num_heads_upsample=-1,
        use_scale_shift_norm=False,
        resblock_updown=False,
        use_new_attention_order=False,
        use_spatial_transformer=False,    # custom transformer support
        transformer_depth=1,              # custom transformer support
        context_dim=None,                 # custom transformer support
        n_embed=None,                     # custom support for prediction of discrete ids into codebook of first stage vq model
        legacy=True,
        disable_self_attentions=None,
        num_attention_blocks=None,
        disable_middle_self_attn=False,
        use_linear_in_transformer=False,
    ):
        super().__init__()
        if use_spatial_transformer:
            assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'

        if context_dim is not None:
            assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
            from omegaconf.listconfig import ListConfig
            if type(context_dim) == ListConfig:
                context_dim = list(context_dim)

        if num_heads_upsample == -1:
            num_heads_upsample = num_heads

        if num_heads == -1:
            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'

        if num_head_channels == -1:
            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'

        self.dims = dims
        self.image_size = image_size
        self.in_channels = in_channels
        self.out_channels = out_channels
        self.model_channels = model_channels
        if isinstance(num_res_blocks, int):
            self.num_res_blocks = len(channel_mult) * [num_res_blocks]
        else:
            if len(num_res_blocks) != len(channel_mult):
                raise ValueError("provide num_res_blocks either as an int (globally constant) or "
                                 "as a list/tuple (per-level) with the same length as channel_mult")
            self.num_res_blocks = num_res_blocks
        if disable_self_attentions is not None:
            # should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
            assert len(disable_self_attentions) == len(channel_mult)
        if num_attention_blocks is not None:
            assert len(num_attention_blocks) == len(self.num_res_blocks)
            assert all(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
            print(f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
                  f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
                  f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
                  f"attention will still not be set.")

        self.attention_resolutions = attention_resolutions
        self.dropout = dropout
        self.channel_mult = channel_mult
        self.conv_resample = conv_resample
        self.use_checkpoint = use_checkpoint
        self.dtype = th.float16 if use_fp16 else th.float32
        self.num_heads = num_heads
        self.num_head_channels = num_head_channels
        self.num_heads_upsample = num_heads_upsample
        self.predict_codebook_ids = n_embed is not None

        time_embed_dim = model_channels * 4
        self.time_embed = nn.Sequential(
            linear(model_channels, time_embed_dim),
            nn.SiLU(),
            linear(time_embed_dim, time_embed_dim),
        )

        self.input_blocks = nn.ModuleList(
            [
                TimestepEmbedSequential(
                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
                )
            ]
        )

        self.input_hint_block = TimestepEmbedSequential(
                    conv_nd(dims, hint_channels, 16, 3, padding=1),
                    nn.SiLU(),
                    conv_nd(dims, 16, 16, 3, padding=1),
                    nn.SiLU(),
                    conv_nd(dims, 16, 32, 3, padding=1, stride=2),
                    nn.SiLU(),
                    conv_nd(dims, 32, 32, 3, padding=1),
                    nn.SiLU(),
                    conv_nd(dims, 32, 96, 3, padding=1, stride=2),
                    nn.SiLU(),
                    conv_nd(dims, 96, 96, 3, padding=1),
                    nn.SiLU(),
                    conv_nd(dims, 96, 256, 3, padding=1, stride=2),
                    nn.SiLU(),
                    zero_module(conv_nd(dims, 256, model_channels, 3, padding=1))
        )

        self._feature_size = model_channels
        input_block_chans = [model_channels]
        ch = model_channels
        ds = 1
        for level, mult in enumerate(channel_mult):
            for nr in range(self.num_res_blocks[level]):
                layers = [
                    ResBlock(
                        ch,
                        time_embed_dim,
                        dropout,
                        out_channels=mult * model_channels,
                        dims=dims,
                        use_checkpoint=use_checkpoint,
                        use_scale_shift_norm=use_scale_shift_norm,
                    )
                ]
                ch = mult * model_channels
                if ds in attention_resolutions:
                    if num_head_channels == -1:
                        dim_head = ch // num_heads
                    else:
                        num_heads = ch // num_head_channels
                        dim_head = num_head_channels
                    if legacy:
                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
                    if exists(disable_self_attentions):
                        disabled_sa = disable_self_attentions[level]
                    else:
                        disabled_sa = False

                    if not exists(num_attention_blocks) or nr < num_attention_blocks[level]:
                        layers.append(
                            AttentionBlock(
                                ch,
                                use_checkpoint=use_checkpoint,
                                num_heads=num_heads,
                                num_head_channels=dim_head,
                                use_new_attention_order=use_new_attention_order,
                            ) if not use_spatial_transformer else SpatialTransformer(
                                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
                                use_checkpoint=use_checkpoint
                            )
                        )
                self.input_blocks.append(TimestepEmbedSequential(*layers))
                self._feature_size += ch
                input_block_chans.append(ch)
            if level != len(channel_mult) - 1:
                out_ch = ch
                self.input_blocks.append(
                    TimestepEmbedSequential(
                        ResBlock(
                            ch,
                            time_embed_dim,
                            dropout,
                            out_channels=out_ch,
                            dims=dims,
                            use_checkpoint=use_checkpoint,
                            use_scale_shift_norm=use_scale_shift_norm,
                            down=True,
                        )
                        if resblock_updown
                        else Downsample(
                            ch, conv_resample, dims=dims, out_channels=out_ch
                        )
                    )
                )
                ch = out_ch
                input_block_chans.append(ch)
                ds *= 2
                self._feature_size += ch

        if num_head_channels == -1:
            dim_head = ch // num_heads
        else:
            num_heads = ch // num_head_channels
            dim_head = num_head_channels
        if legacy:
            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
        self.middle_block = TimestepEmbedSequential(
            ResBlock(
                ch,
                time_embed_dim,
                dropout,
                dims=dims,
                use_checkpoint=use_checkpoint,
                use_scale_shift_norm=use_scale_shift_norm,
            ),
            AttentionBlock(
                ch,
                use_checkpoint=use_checkpoint,
                num_heads=num_heads,
                num_head_channels=dim_head,
                use_new_attention_order=use_new_attention_order,
            ) if not use_spatial_transformer else SpatialTransformer(  # always uses a self-attn
                            ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
                            disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer,
                            use_checkpoint=use_checkpoint
                        ),
            ResBlock(
                ch,
                time_embed_dim,
                dropout,
                dims=dims,
                use_checkpoint=use_checkpoint,
                use_scale_shift_norm=use_scale_shift_norm,
            ),
        )
        self._feature_size += ch


        self.output_blocks = nn.ModuleList([])
        for level, mult in list(enumerate(channel_mult))[::-1]:
            for i in range(self.num_res_blocks[level] + 1):
                ich = input_block_chans.pop()
                layers = [
                    ResBlock(
                        ch + ich,
                        time_embed_dim,
                        dropout,
                        out_channels=model_channels * mult,
                        dims=dims,
                        use_checkpoint=use_checkpoint,
                        use_scale_shift_norm=use_scale_shift_norm,
                    )
                ]
                ch = model_channels * mult
                if ds in attention_resolutions:
                    if num_head_channels == -1:
                        dim_head = ch // num_heads
                    else:
                        num_heads = ch // num_head_channels
                        dim_head = num_head_channels
                    if legacy:
                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
                    if exists(disable_self_attentions):
                        disabled_sa = disable_self_attentions[level]
                    else:
                        disabled_sa = False

                    if not exists(num_attention_blocks) or i < num_attention_blocks[level]:
                        layers.append(
                            AttentionBlock(
                                ch,
                                use_checkpoint=use_checkpoint,
                                num_heads=num_heads_upsample,
                                num_head_channels=dim_head,
                                use_new_attention_order=use_new_attention_order,
                            ) if not use_spatial_transformer else SpatialTransformer(
                                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
                                use_checkpoint=use_checkpoint
                            )
                        )
                if level and i == self.num_res_blocks[level]:
                    out_ch = ch
                    layers.append(
                        ResBlock(
                            ch,
                            time_embed_dim,
                            dropout,
                            out_channels=out_ch,
                            dims=dims,
                            use_checkpoint=use_checkpoint,
                            use_scale_shift_norm=use_scale_shift_norm,
                            up=True,
                        )
                        if resblock_updown
                        else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
                    )
                    ds //= 2
                self.output_blocks.append(TimestepEmbedSequential(*layers))
                self._feature_size += ch


    def forward(self, x, hint, timesteps, context, attention_bank=None, attention_mode=None,uc=False, **kwargs):
        hs = []
        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
        emb = self.time_embed(t_emb)
        banks = attention_bank
        outs = []
        h = x.type(self.dtype)
        for module in self.input_blocks:
            h = module(h, emb, context, banks, attention_mode,uc)
            hs.append(h)
            
        h = self.middle_block(h, emb, context, banks, attention_mode,uc)
        
        for module in self.output_blocks:
            h = th.cat([h, hs.pop()], dim=1)
            h = module(h, emb, context, banks, attention_mode,uc)

        return outs

### ControlNet Origin
class ControlNet(nn.Module):
    def __init__(
        self,
        image_size,
        in_channels,
        model_channels,
        hint_channels,
        num_res_blocks,
        attention_resolutions,
        dropout=0,
        channel_mult=(1, 2, 4, 8),
        conv_resample=True,
        dims=2,
        use_checkpoint=False,
        use_fp16=False,
        num_heads=-1,
        num_head_channels=-1,
        num_heads_upsample=-1,
        use_scale_shift_norm=False,
        resblock_updown=False,
        use_new_attention_order=False,
        use_spatial_transformer=False,    # custom transformer support
        transformer_depth=1,              # custom transformer support
        context_dim=None,                 # custom transformer support
        n_embed=None,                     # custom support for prediction of discrete ids into codebook of first stage vq model
        legacy=True,
        disable_self_attentions=None,
        num_attention_blocks=None,
        disable_middle_self_attn=False,
        use_linear_in_transformer=False,
    ):
        super().__init__()
        if use_spatial_transformer:
            assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'

        if context_dim is not None:
            assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
            from omegaconf.listconfig import ListConfig
            if type(context_dim) == ListConfig:
                context_dim = list(context_dim)

        if num_heads_upsample == -1:
            num_heads_upsample = num_heads

        if num_heads == -1:
            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'

        if num_head_channels == -1:
            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'

        self.dims = dims
        self.image_size = image_size
        self.in_channels = in_channels
        self.model_channels = model_channels
        if isinstance(num_res_blocks, int):
            self.num_res_blocks = len(channel_mult) * [num_res_blocks]
        else:
            if len(num_res_blocks) != len(channel_mult):
                raise ValueError("provide num_res_blocks either as an int (globally constant) or "
                                 "as a list/tuple (per-level) with the same length as channel_mult")
            self.num_res_blocks = num_res_blocks
        if disable_self_attentions is not None:
            # should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
            assert len(disable_self_attentions) == len(channel_mult)
        if num_attention_blocks is not None:
            assert len(num_attention_blocks) == len(self.num_res_blocks)
            assert all(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
            print(f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
                  f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
                  f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
                  f"attention will still not be set.")

        self.attention_resolutions = attention_resolutions
        self.dropout = dropout
        self.channel_mult = channel_mult
        self.conv_resample = conv_resample
        self.use_checkpoint = use_checkpoint
        self.dtype = th.float16 if use_fp16 else th.float32
        self.num_heads = num_heads
        self.num_head_channels = num_head_channels
        self.num_heads_upsample = num_heads_upsample
        self.predict_codebook_ids = n_embed is not None

        time_embed_dim = model_channels * 4
        self.time_embed = nn.Sequential(
            linear(model_channels, time_embed_dim),
            nn.SiLU(),
            linear(time_embed_dim, time_embed_dim),
        )

        self.input_blocks = nn.ModuleList(
            [
                TimestepEmbedSequential(
                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
                )
            ]
        )
        self.zero_convs = nn.ModuleList([self.make_zero_conv(model_channels)])

        self.input_hint_block = TimestepEmbedSequential(
                    conv_nd(dims, hint_channels, 16, 3, padding=1),
                    nn.SiLU(),
                    conv_nd(dims, 16, 16, 3, padding=1),
                    nn.SiLU(),
                    conv_nd(dims, 16, 32, 3, padding=1, stride=2),
                    nn.SiLU(),
                    conv_nd(dims, 32, 32, 3, padding=1),
                    nn.SiLU(),
                    conv_nd(dims, 32, 96, 3, padding=1, stride=2),
                    nn.SiLU(),
                    conv_nd(dims, 96, 96, 3, padding=1),
                    nn.SiLU(),
                    conv_nd(dims, 96, 256, 3, padding=1, stride=2),
                    nn.SiLU(),
                    zero_module(conv_nd(dims, 256, model_channels, 3, padding=1))
        )

        self._feature_size = model_channels
        input_block_chans = [model_channels]
        ch = model_channels
        ds = 1
        for level, mult in enumerate(channel_mult):
            for nr in range(self.num_res_blocks[level]):
                layers = [
                    ResBlock(
                        ch,
                        time_embed_dim,
                        dropout,
                        out_channels=mult * model_channels,
                        dims=dims,
                        use_checkpoint=use_checkpoint,
                        use_scale_shift_norm=use_scale_shift_norm,
                    )
                ]
                ch = mult * model_channels
                if ds in attention_resolutions:
                    if num_head_channels == -1:
                        dim_head = ch // num_heads
                    else:
                        num_heads = ch // num_head_channels
                        dim_head = num_head_channels
                    if legacy:
                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
                    if exists(disable_self_attentions):
                        disabled_sa = disable_self_attentions[level]
                    else:
                        disabled_sa = False

                    if not exists(num_attention_blocks) or nr < num_attention_blocks[level]:
                        layers.append(
                            AttentionBlock(
                                ch,
                                use_checkpoint=use_checkpoint,
                                num_heads=num_heads,
                                num_head_channels=dim_head,
                                use_new_attention_order=use_new_attention_order,
                            ) if not use_spatial_transformer else SpatialTransformer(
                                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
                                use_checkpoint=use_checkpoint
                            )
                        )
                self.input_blocks.append(TimestepEmbedSequential(*layers))
                self.zero_convs.append(self.make_zero_conv(ch))
                self._feature_size += ch
                input_block_chans.append(ch)
            if level != len(channel_mult) - 1:
                out_ch = ch
                self.input_blocks.append(
                    TimestepEmbedSequential(
                        ResBlock(
                            ch,
                            time_embed_dim,
                            dropout,
                            out_channels=out_ch,
                            dims=dims,
                            use_checkpoint=use_checkpoint,
                            use_scale_shift_norm=use_scale_shift_norm,
                            down=True,
                        )
                        if resblock_updown
                        else Downsample(
                            ch, conv_resample, dims=dims, out_channels=out_ch
                        )
                    )
                )
                ch = out_ch
                input_block_chans.append(ch)
                self.zero_convs.append(self.make_zero_conv(ch))
                ds *= 2
                self._feature_size += ch

        if num_head_channels == -1:
            dim_head = ch // num_heads
        else:
            num_heads = ch // num_head_channels
            dim_head = num_head_channels
        if legacy:
            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
        self.middle_block = TimestepEmbedSequential(
            ResBlock(
                ch,
                time_embed_dim,
                dropout,
                dims=dims,
                use_checkpoint=use_checkpoint,
                use_scale_shift_norm=use_scale_shift_norm,
            ),
            AttentionBlock(
                ch,
                use_checkpoint=use_checkpoint,
                num_heads=num_heads,
                num_head_channels=dim_head,
                use_new_attention_order=use_new_attention_order,
            ) if not use_spatial_transformer else SpatialTransformer(  # always uses a self-attn
                            ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
                            disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer,
                            use_checkpoint=use_checkpoint
                        ),
            ResBlock(
                ch,
                time_embed_dim,
                dropout,
                dims=dims,
                use_checkpoint=use_checkpoint,
                use_scale_shift_norm=use_scale_shift_norm,
            ),
        )
        self.middle_block_out = self.make_zero_conv(ch)
        self._feature_size += ch

    def make_zero_conv(self, channels):
        return TimestepEmbedSequential(zero_module(conv_nd(self.dims, channels, channels, 1, padding=0)))

    def forward(self, x, hint, timesteps, context, **kwargs):
        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
        emb = self.time_embed(t_emb)

        guided_hint = self.input_hint_block(hint, emb, context)

        outs = []
        h = x.type(self.dtype)

        for module, zero_conv in zip(self.input_blocks, self.zero_convs):
            if guided_hint is not None:
                h = module(h, emb, context)
                h += guided_hint
                guided_hint = None
            else:
                h = module(h, emb, context)
            outs.append(zero_conv(h, emb, context))

        h = self.middle_block(h, emb, context)
        outs.append(self.middle_block_out(h, emb, context))

        return outs

class ControlLDMReferenceOnly_Temporal_Pose_Local(LatentDiffusionReferenceOnly):

    def __init__(self, control_key, only_mid_control,appearance_control_stage_config, pose_control_stage_config, local_pose_control_stage_config, *args, **kwargs):
        super().__init__(*args, **kwargs)
        print(args)
        print(kwargs)
        self.control_key = control_key
        self.only_mid_control = only_mid_control
        self.control_enabled = True
        self.appearance_control_model = instantiate_from_config(appearance_control_stage_config)
        self.pose_control_model = instantiate_from_config(pose_control_stage_config)
        self.local_pose_control_model = instantiate_from_config(local_pose_control_stage_config)

    def apply_model(self, x_noisy, t, cond, reference_image_noisy, more_reference_image_noisy=[], uc=False,*args, **kwargs):
        assert isinstance(cond, dict)
        diffusion_model = self.model.diffusion_model
        cond_txt = torch.cat(cond['c_crossattn'], 1)
        if self.control_enabled and 'c_crossattn_void' in cond and cond['c_crossattn_void'] is not None:
            cond_txt_void = torch.cat(cond['c_crossattn_void'], 1)
        else:
            cond_txt_void = cond_txt
        attention_bank = []

        if reference_image_noisy is not None:
            empty_outs = self.appearance_control_model(x=reference_image_noisy, hint=None, timesteps=t, context=cond_txt_void, attention_bank=attention_bank, attention_mode='write',uc=uc)
        for m_reference_image_noisy in more_reference_image_noisy:
            l_attention_bank = []
            empty_outs = self.appearance_control_model(x=m_reference_image_noisy, hint=None, timesteps=t, context=cond_txt_void, attention_bank=l_attention_bank, attention_mode='write',uc=uc)
            for j in range(len(attention_bank)):
                for k in range(len(attention_bank[j])):
                    attention_bank[j][k] = torch.concat([attention_bank[j][k], l_attention_bank[j][k]], dim=1)

        if not uc:
            if self.control_enabled and 'c_concat' in cond and cond['c_concat'] is not None:
                cond_hint = torch.cat(cond['c_concat'], 1)
                pose_control = self.pose_control_model(x=x_noisy, hint=cond_hint, timesteps=t, context=cond_txt_void)
            
            if self.control_enabled and 'local_c_concat' in cond and cond['local_c_concat']  is not None:
                cond_hint = torch.cat(cond['local_c_concat'], 1)
                local_pose_control = self.local_pose_control_model(x=x_noisy, hint=cond_hint, timesteps=t, context=cond_txt_void) 
        else:
            pose_control = None
            local_pose_control = None
        eps = diffusion_model(x=x_noisy, timesteps=t, context=cond_txt, control=attention_bank, pose_control=pose_control, local_pose_control=local_pose_control, only_mid_control=self.only_mid_control, attention_mode='read',uc=uc)
        return eps

    @torch.no_grad()
    def get_unconditional_conditioning(self, N):
        return self.get_learned_conditioning([""] * N)