modelscope/DiffSynth-Painter
14
1import torch2from .sd_unet import Timesteps, ResnetBlock, AttentionBlock, PushBlock, DownSampler3from .tiler import TileWorker4 5 6class ControlNetConditioningLayer(torch.nn.Module):7 def __init__(self, channels = (3, 16, 32, 96, 256, 320)):8 super().__init__()9 self.blocks = torch.nn.ModuleList([])10 self.blocks.append(torch.nn.Conv2d(channels[0], channels[1], kernel_size=3, padding=1))11 self.blocks.append(torch.nn.SiLU())12 for i in range(1, len(channels) - 2):13 self.blocks.append(torch.nn.Conv2d(channels[i], channels[i], kernel_size=3, padding=1))14 self.blocks.append(torch.nn.SiLU())15 self.blocks.append(torch.nn.Conv2d(channels[i], channels[i+1], kernel_size=3, padding=1, stride=2))16 self.blocks.append(torch.nn.SiLU())17 self.blocks.append(torch.nn.Conv2d(channels[-2], channels[-1], kernel_size=3, padding=1))18 19 def forward(self, conditioning):20 for block in self.blocks:21 conditioning = block(conditioning)22 return conditioning23 24 25class SDControlNet(torch.nn.Module):26 def __init__(self, global_pool=False):27 super().__init__()28 self.time_proj = Timesteps(320)29 self.time_embedding = torch.nn.Sequential(30 torch.nn.Linear(320, 1280),31 torch.nn.SiLU(),32 torch.nn.Linear(1280, 1280)33 )34 self.conv_in = torch.nn.Conv2d(4, 320, kernel_size=3, padding=1)35 36 self.controlnet_conv_in = ControlNetConditioningLayer(channels=(3, 16, 32, 96, 256, 320))37 38 self.blocks = torch.nn.ModuleList([39 # CrossAttnDownBlock2D40 ResnetBlock(320, 320, 1280),41 AttentionBlock(8, 40, 320, 1, 768),42 PushBlock(),43 ResnetBlock(320, 320, 1280),44 AttentionBlock(8, 40, 320, 1, 768),45 PushBlock(),46 DownSampler(320),47 PushBlock(),48 # CrossAttnDownBlock2D49 ResnetBlock(320, 640, 1280),50 AttentionBlock(8, 80, 640, 1, 768),51 PushBlock(),52 ResnetBlock(640, 640, 1280),53 AttentionBlock(8, 80, 640, 1, 768),54 PushBlock(),55 DownSampler(640),56 PushBlock(),57 # CrossAttnDownBlock2D58 ResnetBlock(640, 1280, 1280),59 AttentionBlock(8, 160, 1280, 1, 768),60 PushBlock(),61 ResnetBlock(1280, 1280, 1280),62 AttentionBlock(8, 160, 1280, 1, 768),63 PushBlock(),64 DownSampler(1280),65 PushBlock(),66 # DownBlock2D67 ResnetBlock(1280, 1280, 1280),68 PushBlock(),69 ResnetBlock(1280, 1280, 1280),70 PushBlock(),71 # UNetMidBlock2DCrossAttn72 ResnetBlock(1280, 1280, 1280),73 AttentionBlock(8, 160, 1280, 1, 768),74 ResnetBlock(1280, 1280, 1280),75 PushBlock()76 ])77 78 self.controlnet_blocks = torch.nn.ModuleList([79 torch.nn.Conv2d(320, 320, kernel_size=(1, 1)),80 torch.nn.Conv2d(320, 320, kernel_size=(1, 1), bias=False),81 torch.nn.Conv2d(320, 320, kernel_size=(1, 1), bias=False),82 torch.nn.Conv2d(320, 320, kernel_size=(1, 1), bias=False),83 torch.nn.Conv2d(640, 640, kernel_size=(1, 1)),84 torch.nn.Conv2d(640, 640, kernel_size=(1, 1), bias=False),85 torch.nn.Conv2d(640, 640, kernel_size=(1, 1), bias=False),86 torch.nn.Conv2d(1280, 1280, kernel_size=(1, 1)),87 torch.nn.Conv2d(1280, 1280, kernel_size=(1, 1), bias=False),88 torch.nn.Conv2d(1280, 1280, kernel_size=(1, 1), bias=False),89 torch.nn.Conv2d(1280, 1280, kernel_size=(1, 1), bias=False),90 torch.nn.Conv2d(1280, 1280, kernel_size=(1, 1), bias=False),91 torch.nn.Conv2d(1280, 1280, kernel_size=(1, 1), bias=False),92 ])93 94 self.global_pool = global_pool95 96 def forward(97 self,98 sample, timestep, encoder_hidden_states, conditioning,99 tiled=False, tile_size=64, tile_stride=32,100 **kwargs101 ):102 # 1. time103 time_emb = self.time_proj(timestep).to(sample.dtype)104 time_emb = self.time_embedding(time_emb)105 time_emb = time_emb.repeat(sample.shape[0], 1)106 107 # 2. pre-process108 height, width = sample.shape[2], sample.shape[3]109 hidden_states = self.conv_in(sample) + self.controlnet_conv_in(conditioning)110 text_emb = encoder_hidden_states111 res_stack = [hidden_states]112 113 # 3. blocks114 for i, block in enumerate(self.blocks):115 if tiled and not isinstance(block, PushBlock):116 _, _, inter_height, _ = hidden_states.shape117 resize_scale = inter_height / height118 hidden_states = TileWorker().tiled_forward(119 lambda x: block(x, time_emb, text_emb, res_stack)[0],120 hidden_states,121 int(tile_size * resize_scale),122 int(tile_stride * resize_scale),123 tile_device=hidden_states.device,124 tile_dtype=hidden_states.dtype125 )126 else:127 hidden_states, _, _, _ = block(hidden_states, time_emb, text_emb, res_stack)128 129 # 4. ControlNet blocks130 controlnet_res_stack = [block(res) for block, res in zip(self.controlnet_blocks, res_stack)]131 132 # pool133 if self.global_pool:134 controlnet_res_stack = [res.mean(dim=(2, 3), keepdim=True) for res in controlnet_res_stack]135 136 return controlnet_res_stack137 138 @staticmethod139 def state_dict_converter():140 return SDControlNetStateDictConverter()141 142 143class SDControlNetStateDictConverter:144 def __init__(self):145 pass146 147 def from_diffusers(self, state_dict):148 # architecture149 block_types = [150 'ResnetBlock', 'AttentionBlock', 'PushBlock', 'ResnetBlock', 'AttentionBlock', 'PushBlock', 'DownSampler', 'PushBlock',151 'ResnetBlock', 'AttentionBlock', 'PushBlock', 'ResnetBlock', 'AttentionBlock', 'PushBlock', 'DownSampler', 'PushBlock',152 'ResnetBlock', 'AttentionBlock', 'PushBlock', 'ResnetBlock', 'AttentionBlock', 'PushBlock', 'DownSampler', 'PushBlock',153 'ResnetBlock', 'PushBlock', 'ResnetBlock', 'PushBlock', 154 'ResnetBlock', 'AttentionBlock', 'ResnetBlock',155 'PopBlock', 'ResnetBlock', 'PopBlock', 'ResnetBlock', 'PopBlock', 'ResnetBlock', 'UpSampler',156 'PopBlock', 'ResnetBlock', 'AttentionBlock', 'PopBlock', 'ResnetBlock', 'AttentionBlock', 'PopBlock', 'ResnetBlock', 'AttentionBlock', 'UpSampler',157 'PopBlock', 'ResnetBlock', 'AttentionBlock', 'PopBlock', 'ResnetBlock', 'AttentionBlock', 'PopBlock', 'ResnetBlock', 'AttentionBlock', 'UpSampler',158 'PopBlock', 'ResnetBlock', 'AttentionBlock', 'PopBlock', 'ResnetBlock', 'AttentionBlock', 'PopBlock', 'ResnetBlock', 'AttentionBlock'159 ]160 161 # controlnet_rename_dict162 controlnet_rename_dict = {163 "controlnet_cond_embedding.conv_in.weight": "controlnet_conv_in.blocks.0.weight",164 "controlnet_cond_embedding.conv_in.bias": "controlnet_conv_in.blocks.0.bias",165 "controlnet_cond_embedding.blocks.0.weight": "controlnet_conv_in.blocks.2.weight",166 "controlnet_cond_embedding.blocks.0.bias": "controlnet_conv_in.blocks.2.bias",167 "controlnet_cond_embedding.blocks.1.weight": "controlnet_conv_in.blocks.4.weight",168 "controlnet_cond_embedding.blocks.1.bias": "controlnet_conv_in.blocks.4.bias",169 "controlnet_cond_embedding.blocks.2.weight": "controlnet_conv_in.blocks.6.weight",170 "controlnet_cond_embedding.blocks.2.bias": "controlnet_conv_in.blocks.6.bias",171 "controlnet_cond_embedding.blocks.3.weight": "controlnet_conv_in.blocks.8.weight",172 "controlnet_cond_embedding.blocks.3.bias": "controlnet_conv_in.blocks.8.bias",173 "controlnet_cond_embedding.blocks.4.weight": "controlnet_conv_in.blocks.10.weight",174 "controlnet_cond_embedding.blocks.4.bias": "controlnet_conv_in.blocks.10.bias",175 "controlnet_cond_embedding.blocks.5.weight": "controlnet_conv_in.blocks.12.weight",176 "controlnet_cond_embedding.blocks.5.bias": "controlnet_conv_in.blocks.12.bias",177 "controlnet_cond_embedding.conv_out.weight": "controlnet_conv_in.blocks.14.weight",178 "controlnet_cond_embedding.conv_out.bias": "controlnet_conv_in.blocks.14.bias",179 }180 181 # Rename each parameter182 name_list = sorted([name for name in state_dict])183 rename_dict = {}184 block_id = {"ResnetBlock": -1, "AttentionBlock": -1, "DownSampler": -1, "UpSampler": -1}185 last_block_type_with_id = {"ResnetBlock": "", "AttentionBlock": "", "DownSampler": "", "UpSampler": ""}186 for name in name_list:187 names = name.split(".")188 if names[0] in ["conv_in", "conv_norm_out", "conv_out"]:189 pass190 elif name in controlnet_rename_dict:191 names = controlnet_rename_dict[name].split(".")192 elif names[0] == "controlnet_down_blocks":193 names[0] = "controlnet_blocks"194 elif names[0] == "controlnet_mid_block":195 names = ["controlnet_blocks", "12", names[-1]]196 elif names[0] in ["time_embedding", "add_embedding"]:197 if names[0] == "add_embedding":198 names[0] = "add_time_embedding"199 names[1] = {"linear_1": "0", "linear_2": "2"}[names[1]]200 elif names[0] in ["down_blocks", "mid_block", "up_blocks"]:201 if names[0] == "mid_block":202 names.insert(1, "0")203 block_type = {"resnets": "ResnetBlock", "attentions": "AttentionBlock", "downsamplers": "DownSampler", "upsamplers": "UpSampler"}[names[2]]204 block_type_with_id = ".".join(names[:4])205 if block_type_with_id != last_block_type_with_id[block_type]:206 block_id[block_type] += 1207 last_block_type_with_id[block_type] = block_type_with_id208 while block_id[block_type] < len(block_types) and block_types[block_id[block_type]] != block_type:209 block_id[block_type] += 1210 block_type_with_id = ".".join(names[:4])211 names = ["blocks", str(block_id[block_type])] + names[4:]212 if "ff" in names:213 ff_index = names.index("ff")214 component = ".".join(names[ff_index:ff_index+3])215 component = {"ff.net.0": "act_fn", "ff.net.2": "ff"}[component]216 names = names[:ff_index] + [component] + names[ff_index+3:]217 if "to_out" in names:218 names.pop(names.index("to_out") + 1)219 else:220 raise ValueError(f"Unknown parameters: {name}")221 rename_dict[name] = ".".join(names)222 223 # Convert state_dict224 state_dict_ = {}225 for name, param in state_dict.items():226 if ".proj_in." in name or ".proj_out." in name:227 param = param.squeeze()228 if rename_dict[name] in [229 "controlnet_blocks.1.bias", "controlnet_blocks.2.bias", "controlnet_blocks.3.bias", "controlnet_blocks.5.bias", "controlnet_blocks.6.bias",230 "controlnet_blocks.8.bias", "controlnet_blocks.9.bias", "controlnet_blocks.10.bias", "controlnet_blocks.11.bias", "controlnet_blocks.12.bias"231 ]:232 continue233 state_dict_[rename_dict[name]] = param234 return state_dict_235 236 def from_civitai(self, state_dict):237 if "mid_block.resnets.1.time_emb_proj.weight" in state_dict:238 # For controlnets in diffusers format239 return self.from_diffusers(state_dict)240 rename_dict = {241 "control_model.time_embed.0.weight": "time_embedding.0.weight",242 "control_model.time_embed.0.bias": "time_embedding.0.bias",243 "control_model.time_embed.2.weight": "time_embedding.2.weight",244 "control_model.time_embed.2.bias": "time_embedding.2.bias",245 "control_model.input_blocks.0.0.weight": "conv_in.weight",246 "control_model.input_blocks.0.0.bias": "conv_in.bias",247 "control_model.input_blocks.1.0.in_layers.0.weight": "blocks.0.norm1.weight",248 "control_model.input_blocks.1.0.in_layers.0.bias": "blocks.0.norm1.bias",249 "control_model.input_blocks.1.0.in_layers.2.weight": "blocks.0.conv1.weight",250 "control_model.input_blocks.1.0.in_layers.2.bias": "blocks.0.conv1.bias",251 "control_model.input_blocks.1.0.emb_layers.1.weight": "blocks.0.time_emb_proj.weight",252 "control_model.input_blocks.1.0.emb_layers.1.bias": "blocks.0.time_emb_proj.bias",253 "control_model.input_blocks.1.0.out_layers.0.weight": "blocks.0.norm2.weight",254 "control_model.input_blocks.1.0.out_layers.0.bias": "blocks.0.norm2.bias",255 "control_model.input_blocks.1.0.out_layers.3.weight": "blocks.0.conv2.weight",256 "control_model.input_blocks.1.0.out_layers.3.bias": "blocks.0.conv2.bias",257 "control_model.input_blocks.1.1.norm.weight": "blocks.1.norm.weight",258 "control_model.input_blocks.1.1.norm.bias": "blocks.1.norm.bias",259 "control_model.input_blocks.1.1.proj_in.weight": "blocks.1.proj_in.weight",260 "control_model.input_blocks.1.1.proj_in.bias": "blocks.1.proj_in.bias",261 "control_model.input_blocks.1.1.transformer_blocks.0.attn1.to_q.weight": "blocks.1.transformer_blocks.0.attn1.to_q.weight",262 "control_model.input_blocks.1.1.transformer_blocks.0.attn1.to_k.weight": "blocks.1.transformer_blocks.0.attn1.to_k.weight",263 "control_model.input_blocks.1.1.transformer_blocks.0.attn1.to_v.weight": "blocks.1.transformer_blocks.0.attn1.to_v.weight",264 "control_model.input_blocks.1.1.transformer_blocks.0.attn1.to_out.0.weight": "blocks.1.transformer_blocks.0.attn1.to_out.weight",265 "control_model.input_blocks.1.1.transformer_blocks.0.attn1.to_out.0.bias": "blocks.1.transformer_blocks.0.attn1.to_out.bias",266 "control_model.input_blocks.1.1.transformer_blocks.0.ff.net.0.proj.weight": "blocks.1.transformer_blocks.0.act_fn.proj.weight",267 "control_model.input_blocks.1.1.transformer_blocks.0.ff.net.0.proj.bias": "blocks.1.transformer_blocks.0.act_fn.proj.bias",268 "control_model.input_blocks.1.1.transformer_blocks.0.ff.net.2.weight": "blocks.1.transformer_blocks.0.ff.weight",269 "control_model.input_blocks.1.1.transformer_blocks.0.ff.net.2.bias": "blocks.1.transformer_blocks.0.ff.bias",270 "control_model.input_blocks.1.1.transformer_blocks.0.attn2.to_q.weight": "blocks.1.transformer_blocks.0.attn2.to_q.weight",271 "control_model.input_blocks.1.1.transformer_blocks.0.attn2.to_k.weight": "blocks.1.transformer_blocks.0.attn2.to_k.weight",272 "control_model.input_blocks.1.1.transformer_blocks.0.attn2.to_v.weight": "blocks.1.transformer_blocks.0.attn2.to_v.weight",273 "control_model.input_blocks.1.1.transformer_blocks.0.attn2.to_out.0.weight": "blocks.1.transformer_blocks.0.attn2.to_out.weight",274 "control_model.input_blocks.1.1.transformer_blocks.0.attn2.to_out.0.bias": "blocks.1.transformer_blocks.0.attn2.to_out.bias",275 "control_model.input_blocks.1.1.transformer_blocks.0.norm1.weight": "blocks.1.transformer_blocks.0.norm1.weight",276 "control_model.input_blocks.1.1.transformer_blocks.0.norm1.bias": "blocks.1.transformer_blocks.0.norm1.bias",277 "control_model.input_blocks.1.1.transformer_blocks.0.norm2.weight": "blocks.1.transformer_blocks.0.norm2.weight",278 "control_model.input_blocks.1.1.transformer_blocks.0.norm2.bias": "blocks.1.transformer_blocks.0.norm2.bias",279 "control_model.input_blocks.1.1.transformer_blocks.0.norm3.weight": "blocks.1.transformer_blocks.0.norm3.weight",280 "control_model.input_blocks.1.1.transformer_blocks.0.norm3.bias": "blocks.1.transformer_blocks.0.norm3.bias",281 "control_model.input_blocks.1.1.proj_out.weight": "blocks.1.proj_out.weight",282 "control_model.input_blocks.1.1.proj_out.bias": "blocks.1.proj_out.bias",283 "control_model.input_blocks.2.0.in_layers.0.weight": "blocks.3.norm1.weight",284 "control_model.input_blocks.2.0.in_layers.0.bias": "blocks.3.norm1.bias",285 "control_model.input_blocks.2.0.in_layers.2.weight": "blocks.3.conv1.weight",286 "control_model.input_blocks.2.0.in_layers.2.bias": "blocks.3.conv1.bias",287 "control_model.input_blocks.2.0.emb_layers.1.weight": "blocks.3.time_emb_proj.weight",288 "control_model.input_blocks.2.0.emb_layers.1.bias": "blocks.3.time_emb_proj.bias",289 "control_model.input_blocks.2.0.out_layers.0.weight": "blocks.3.norm2.weight",290 "control_model.input_blocks.2.0.out_layers.0.bias": "blocks.3.norm2.bias",291 "control_model.input_blocks.2.0.out_layers.3.weight": "blocks.3.conv2.weight",292 "control_model.input_blocks.2.0.out_layers.3.bias": "blocks.3.conv2.bias",293 "control_model.input_blocks.2.1.norm.weight": "blocks.4.norm.weight",294 "control_model.input_blocks.2.1.norm.bias": "blocks.4.norm.bias",295 "control_model.input_blocks.2.1.proj_in.weight": "blocks.4.proj_in.weight",296 "control_model.input_blocks.2.1.proj_in.bias": "blocks.4.proj_in.bias",297 "control_model.input_blocks.2.1.transformer_blocks.0.attn1.to_q.weight": "blocks.4.transformer_blocks.0.attn1.to_q.weight",298 "control_model.input_blocks.2.1.transformer_blocks.0.attn1.to_k.weight": "blocks.4.transformer_blocks.0.attn1.to_k.weight",299 "control_model.input_blocks.2.1.transformer_blocks.0.attn1.to_v.weight": "blocks.4.transformer_blocks.0.attn1.to_v.weight",300 "control_model.input_blocks.2.1.transformer_blocks.0.attn1.to_out.0.weight": "blocks.4.transformer_blocks.0.attn1.to_out.weight",301 "control_model.input_blocks.2.1.transformer_blocks.0.attn1.to_out.0.bias": "blocks.4.transformer_blocks.0.attn1.to_out.bias",302 "control_model.input_blocks.2.1.transformer_blocks.0.ff.net.0.proj.weight": "blocks.4.transformer_blocks.0.act_fn.proj.weight",303 "control_model.input_blocks.2.1.transformer_blocks.0.ff.net.0.proj.bias": "blocks.4.transformer_blocks.0.act_fn.proj.bias",304 "control_model.input_blocks.2.1.transformer_blocks.0.ff.net.2.weight": "blocks.4.transformer_blocks.0.ff.weight",305 "control_model.input_blocks.2.1.transformer_blocks.0.ff.net.2.bias": "blocks.4.transformer_blocks.0.ff.bias",306 "control_model.input_blocks.2.1.transformer_blocks.0.attn2.to_q.weight": "blocks.4.transformer_blocks.0.attn2.to_q.weight",307 "control_model.input_blocks.2.1.transformer_blocks.0.attn2.to_k.weight": "blocks.4.transformer_blocks.0.attn2.to_k.weight",308 "control_model.input_blocks.2.1.transformer_blocks.0.attn2.to_v.weight": "blocks.4.transformer_blocks.0.attn2.to_v.weight",309 "control_model.input_blocks.2.1.transformer_blocks.0.attn2.to_out.0.weight": "blocks.4.transformer_blocks.0.attn2.to_out.weight",310 "control_model.input_blocks.2.1.transformer_blocks.0.attn2.to_out.0.bias": "blocks.4.transformer_blocks.0.attn2.to_out.bias",311 "control_model.input_blocks.2.1.transformer_blocks.0.norm1.weight": "blocks.4.transformer_blocks.0.norm1.weight",312 "control_model.input_blocks.2.1.transformer_blocks.0.norm1.bias": "blocks.4.transformer_blocks.0.norm1.bias",313 "control_model.input_blocks.2.1.transformer_blocks.0.norm2.weight": "blocks.4.transformer_blocks.0.norm2.weight",314 "control_model.input_blocks.2.1.transformer_blocks.0.norm2.bias": "blocks.4.transformer_blocks.0.norm2.bias",315 "control_model.input_blocks.2.1.transformer_blocks.0.norm3.weight": "blocks.4.transformer_blocks.0.norm3.weight",316 "control_model.input_blocks.2.1.transformer_blocks.0.norm3.bias": "blocks.4.transformer_blocks.0.norm3.bias",317 "control_model.input_blocks.2.1.proj_out.weight": "blocks.4.proj_out.weight",318 "control_model.input_blocks.2.1.proj_out.bias": "blocks.4.proj_out.bias",319 "control_model.input_blocks.3.0.op.weight": "blocks.6.conv.weight",320 "control_model.input_blocks.3.0.op.bias": "blocks.6.conv.bias",321 "control_model.input_blocks.4.0.in_layers.0.weight": "blocks.8.norm1.weight",322 "control_model.input_blocks.4.0.in_layers.0.bias": "blocks.8.norm1.bias",323 "control_model.input_blocks.4.0.in_layers.2.weight": "blocks.8.conv1.weight",324 "control_model.input_blocks.4.0.in_layers.2.bias": "blocks.8.conv1.bias",325 "control_model.input_blocks.4.0.emb_layers.1.weight": "blocks.8.time_emb_proj.weight",326 "control_model.input_blocks.4.0.emb_layers.1.bias": "blocks.8.time_emb_proj.bias",327 "control_model.input_blocks.4.0.out_layers.0.weight": "blocks.8.norm2.weight",328 "control_model.input_blocks.4.0.out_layers.0.bias": "blocks.8.norm2.bias",329 "control_model.input_blocks.4.0.out_layers.3.weight": "blocks.8.conv2.weight",330 "control_model.input_blocks.4.0.out_layers.3.bias": "blocks.8.conv2.bias",331 "control_model.input_blocks.4.0.skip_connection.weight": "blocks.8.conv_shortcut.weight",332 "control_model.input_blocks.4.0.skip_connection.bias": "blocks.8.conv_shortcut.bias",333 "control_model.input_blocks.4.1.norm.weight": "blocks.9.norm.weight",334 "control_model.input_blocks.4.1.norm.bias": "blocks.9.norm.bias",335 "control_model.input_blocks.4.1.proj_in.weight": "blocks.9.proj_in.weight",336 "control_model.input_blocks.4.1.proj_in.bias": "blocks.9.proj_in.bias",337 "control_model.input_blocks.4.1.transformer_blocks.0.attn1.to_q.weight": "blocks.9.transformer_blocks.0.attn1.to_q.weight",338 "control_model.input_blocks.4.1.transformer_blocks.0.attn1.to_k.weight": "blocks.9.transformer_blocks.0.attn1.to_k.weight",339 "control_model.input_blocks.4.1.transformer_blocks.0.attn1.to_v.weight": "blocks.9.transformer_blocks.0.attn1.to_v.weight",340 "control_model.input_blocks.4.1.transformer_blocks.0.attn1.to_out.0.weight": "blocks.9.transformer_blocks.0.attn1.to_out.weight",341 "control_model.input_blocks.4.1.transformer_blocks.0.attn1.to_out.0.bias": "blocks.9.transformer_blocks.0.attn1.to_out.bias",342 "control_model.input_blocks.4.1.transformer_blocks.0.ff.net.0.proj.weight": "blocks.9.transformer_blocks.0.act_fn.proj.weight",343 "control_model.input_blocks.4.1.transformer_blocks.0.ff.net.0.proj.bias": "blocks.9.transformer_blocks.0.act_fn.proj.bias",344 "control_model.input_blocks.4.1.transformer_blocks.0.ff.net.2.weight": "blocks.9.transformer_blocks.0.ff.weight",345 "control_model.input_blocks.4.1.transformer_blocks.0.ff.net.2.bias": "blocks.9.transformer_blocks.0.ff.bias",346 "control_model.input_blocks.4.1.transformer_blocks.0.attn2.to_q.weight": "blocks.9.transformer_blocks.0.attn2.to_q.weight",347 "control_model.input_blocks.4.1.transformer_blocks.0.attn2.to_k.weight": "blocks.9.transformer_blocks.0.attn2.to_k.weight",348 "control_model.input_blocks.4.1.transformer_blocks.0.attn2.to_v.weight": "blocks.9.transformer_blocks.0.attn2.to_v.weight",349 "control_model.input_blocks.4.1.transformer_blocks.0.attn2.to_out.0.weight": "blocks.9.transformer_blocks.0.attn2.to_out.weight",350 "control_model.input_blocks.4.1.transformer_blocks.0.attn2.to_out.0.bias": "blocks.9.transformer_blocks.0.attn2.to_out.bias",351 "control_model.input_blocks.4.1.transformer_blocks.0.norm1.weight": "blocks.9.transformer_blocks.0.norm1.weight",352 "control_model.input_blocks.4.1.transformer_blocks.0.norm1.bias": "blocks.9.transformer_blocks.0.norm1.bias",353 "control_model.input_blocks.4.1.transformer_blocks.0.norm2.weight": "blocks.9.transformer_blocks.0.norm2.weight",354 "control_model.input_blocks.4.1.transformer_blocks.0.norm2.bias": "blocks.9.transformer_blocks.0.norm2.bias",355 "control_model.input_blocks.4.1.transformer_blocks.0.norm3.weight": "blocks.9.transformer_blocks.0.norm3.weight",356 "control_model.input_blocks.4.1.transformer_blocks.0.norm3.bias": "blocks.9.transformer_blocks.0.norm3.bias",357 "control_model.input_blocks.4.1.proj_out.weight": "blocks.9.proj_out.weight",358 "control_model.input_blocks.4.1.proj_out.bias": "blocks.9.proj_out.bias",359 "control_model.input_blocks.5.0.in_layers.0.weight": "blocks.11.norm1.weight",360 "control_model.input_blocks.5.0.in_layers.0.bias": "blocks.11.norm1.bias",361 "control_model.input_blocks.5.0.in_layers.2.weight": "blocks.11.conv1.weight",362 "control_model.input_blocks.5.0.in_layers.2.bias": "blocks.11.conv1.bias",363 "control_model.input_blocks.5.0.emb_layers.1.weight": "blocks.11.time_emb_proj.weight",364 "control_model.input_blocks.5.0.emb_layers.1.bias": "blocks.11.time_emb_proj.bias",365 "control_model.input_blocks.5.0.out_layers.0.weight": "blocks.11.norm2.weight",366 "control_model.input_blocks.5.0.out_layers.0.bias": "blocks.11.norm2.bias",367 "control_model.input_blocks.5.0.out_layers.3.weight": "blocks.11.conv2.weight",368 "control_model.input_blocks.5.0.out_layers.3.bias": "blocks.11.conv2.bias",369 "control_model.input_blocks.5.1.norm.weight": "blocks.12.norm.weight",370 "control_model.input_blocks.5.1.norm.bias": "blocks.12.norm.bias",371 "control_model.input_blocks.5.1.proj_in.weight": "blocks.12.proj_in.weight",372 "control_model.input_blocks.5.1.proj_in.bias": "blocks.12.proj_in.bias",373 "control_model.input_blocks.5.1.transformer_blocks.0.attn1.to_q.weight": "blocks.12.transformer_blocks.0.attn1.to_q.weight",374 "control_model.input_blocks.5.1.transformer_blocks.0.attn1.to_k.weight": "blocks.12.transformer_blocks.0.attn1.to_k.weight",375 "control_model.input_blocks.5.1.transformer_blocks.0.attn1.to_v.weight": "blocks.12.transformer_blocks.0.attn1.to_v.weight",376 "control_model.input_blocks.5.1.transformer_blocks.0.attn1.to_out.0.weight": "blocks.12.transformer_blocks.0.attn1.to_out.weight",377 "control_model.input_blocks.5.1.transformer_blocks.0.attn1.to_out.0.bias": "blocks.12.transformer_blocks.0.attn1.to_out.bias",378 "control_model.input_blocks.5.1.transformer_blocks.0.ff.net.0.proj.weight": "blocks.12.transformer_blocks.0.act_fn.proj.weight",379 "control_model.input_blocks.5.1.transformer_blocks.0.ff.net.0.proj.bias": "blocks.12.transformer_blocks.0.act_fn.proj.bias",380 "control_model.input_blocks.5.1.transformer_blocks.0.ff.net.2.weight": "blocks.12.transformer_blocks.0.ff.weight",381 "control_model.input_blocks.5.1.transformer_blocks.0.ff.net.2.bias": "blocks.12.transformer_blocks.0.ff.bias",382 "control_model.input_blocks.5.1.transformer_blocks.0.attn2.to_q.weight": "blocks.12.transformer_blocks.0.attn2.to_q.weight",383 "control_model.input_blocks.5.1.transformer_blocks.0.attn2.to_k.weight": "blocks.12.transformer_blocks.0.attn2.to_k.weight",384 "control_model.input_blocks.5.1.transformer_blocks.0.attn2.to_v.weight": "blocks.12.transformer_blocks.0.attn2.to_v.weight",385 "control_model.input_blocks.5.1.transformer_blocks.0.attn2.to_out.0.weight": "blocks.12.transformer_blocks.0.attn2.to_out.weight",386 "control_model.input_blocks.5.1.transformer_blocks.0.attn2.to_out.0.bias": "blocks.12.transformer_blocks.0.attn2.to_out.bias",387 "control_model.input_blocks.5.1.transformer_blocks.0.norm1.weight": "blocks.12.transformer_blocks.0.norm1.weight",388 "control_model.input_blocks.5.1.transformer_blocks.0.norm1.bias": "blocks.12.transformer_blocks.0.norm1.bias",389 "control_model.input_blocks.5.1.transformer_blocks.0.norm2.weight": "blocks.12.transformer_blocks.0.norm2.weight",390 "control_model.input_blocks.5.1.transformer_blocks.0.norm2.bias": "blocks.12.transformer_blocks.0.norm2.bias",391 "control_model.input_blocks.5.1.transformer_blocks.0.norm3.weight": "blocks.12.transformer_blocks.0.norm3.weight",392 "control_model.input_blocks.5.1.transformer_blocks.0.norm3.bias": "blocks.12.transformer_blocks.0.norm3.bias",393 "control_model.input_blocks.5.1.proj_out.weight": "blocks.12.proj_out.weight",394 "control_model.input_blocks.5.1.proj_out.bias": "blocks.12.proj_out.bias",395 "control_model.input_blocks.6.0.op.weight": "blocks.14.conv.weight",396 "control_model.input_blocks.6.0.op.bias": "blocks.14.conv.bias",397 "control_model.input_blocks.7.0.in_layers.0.weight": "blocks.16.norm1.weight",398 "control_model.input_blocks.7.0.in_layers.0.bias": "blocks.16.norm1.bias",399 "control_model.input_blocks.7.0.in_layers.2.weight": "blocks.16.conv1.weight",400 "control_model.input_blocks.7.0.in_layers.2.bias": "blocks.16.conv1.bias",401 "control_model.input_blocks.7.0.emb_layers.1.weight": "blocks.16.time_emb_proj.weight",402 "control_model.input_blocks.7.0.emb_layers.1.bias": "blocks.16.time_emb_proj.bias",403 "control_model.input_blocks.7.0.out_layers.0.weight": "blocks.16.norm2.weight",404 "control_model.input_blocks.7.0.out_layers.0.bias": "blocks.16.norm2.bias",405 "control_model.input_blocks.7.0.out_layers.3.weight": "blocks.16.conv2.weight",406 "control_model.input_blocks.7.0.out_layers.3.bias": "blocks.16.conv2.bias",407 "control_model.input_blocks.7.0.skip_connection.weight": "blocks.16.conv_shortcut.weight",408 "control_model.input_blocks.7.0.skip_connection.bias": "blocks.16.conv_shortcut.bias",409 "control_model.input_blocks.7.1.norm.weight": "blocks.17.norm.weight",410 "control_model.input_blocks.7.1.norm.bias": "blocks.17.norm.bias",411 "control_model.input_blocks.7.1.proj_in.weight": "blocks.17.proj_in.weight",412 "control_model.input_blocks.7.1.proj_in.bias": "blocks.17.proj_in.bias",413 "control_model.input_blocks.7.1.transformer_blocks.0.attn1.to_q.weight": "blocks.17.transformer_blocks.0.attn1.to_q.weight",414 "control_model.input_blocks.7.1.transformer_blocks.0.attn1.to_k.weight": "blocks.17.transformer_blocks.0.attn1.to_k.weight",415 "control_model.input_blocks.7.1.transformer_blocks.0.attn1.to_v.weight": "blocks.17.transformer_blocks.0.attn1.to_v.weight",416 "control_model.input_blocks.7.1.transformer_blocks.0.attn1.to_out.0.weight": "blocks.17.transformer_blocks.0.attn1.to_out.weight",417 "control_model.input_blocks.7.1.transformer_blocks.0.attn1.to_out.0.bias": "blocks.17.transformer_blocks.0.attn1.to_out.bias",418 "control_model.input_blocks.7.1.transformer_blocks.0.ff.net.0.proj.weight": "blocks.17.transformer_blocks.0.act_fn.proj.weight",419 "control_model.input_blocks.7.1.transformer_blocks.0.ff.net.0.proj.bias": "blocks.17.transformer_blocks.0.act_fn.proj.bias",420 "control_model.input_blocks.7.1.transformer_blocks.0.ff.net.2.weight": "blocks.17.transformer_blocks.0.ff.weight",421 "control_model.input_blocks.7.1.transformer_blocks.0.ff.net.2.bias": "blocks.17.transformer_blocks.0.ff.bias",422 "control_model.input_blocks.7.1.transformer_blocks.0.attn2.to_q.weight": "blocks.17.transformer_blocks.0.attn2.to_q.weight",423 "control_model.input_blocks.7.1.transformer_blocks.0.attn2.to_k.weight": "blocks.17.transformer_blocks.0.attn2.to_k.weight",424 "control_model.input_blocks.7.1.transformer_blocks.0.attn2.to_v.weight": "blocks.17.transformer_blocks.0.attn2.to_v.weight",425 "control_model.input_blocks.7.1.transformer_blocks.0.attn2.to_out.0.weight": "blocks.17.transformer_blocks.0.attn2.to_out.weight",426 "control_model.input_blocks.7.1.transformer_blocks.0.attn2.to_out.0.bias": "blocks.17.transformer_blocks.0.attn2.to_out.bias",427 "control_model.input_blocks.7.1.transformer_blocks.0.norm1.weight": "blocks.17.transformer_blocks.0.norm1.weight",428 "control_model.input_blocks.7.1.transformer_blocks.0.norm1.bias": "blocks.17.transformer_blocks.0.norm1.bias",429 "control_model.input_blocks.7.1.transformer_blocks.0.norm2.weight": "blocks.17.transformer_blocks.0.norm2.weight",430 "control_model.input_blocks.7.1.transformer_blocks.0.norm2.bias": "blocks.17.transformer_blocks.0.norm2.bias",431 "control_model.input_blocks.7.1.transformer_blocks.0.norm3.weight": "blocks.17.transformer_blocks.0.norm3.weight",432 "control_model.input_blocks.7.1.transformer_blocks.0.norm3.bias": "blocks.17.transformer_blocks.0.norm3.bias",433 "control_model.input_blocks.7.1.proj_out.weight": "blocks.17.proj_out.weight",434 "control_model.input_blocks.7.1.proj_out.bias": "blocks.17.proj_out.bias",435 "control_model.input_blocks.8.0.in_layers.0.weight": "blocks.19.norm1.weight",436 "control_model.input_blocks.8.0.in_layers.0.bias": "blocks.19.norm1.bias",437 "control_model.input_blocks.8.0.in_layers.2.weight": "blocks.19.conv1.weight",438 "control_model.input_blocks.8.0.in_layers.2.bias": "blocks.19.conv1.bias",439 "control_model.input_blocks.8.0.emb_layers.1.weight": "blocks.19.time_emb_proj.weight",440 "control_model.input_blocks.8.0.emb_layers.1.bias": "blocks.19.time_emb_proj.bias",441 "control_model.input_blocks.8.0.out_layers.0.weight": "blocks.19.norm2.weight",442 "control_model.input_blocks.8.0.out_layers.0.bias": "blocks.19.norm2.bias",443 "control_model.input_blocks.8.0.out_layers.3.weight": "blocks.19.conv2.weight",444 "control_model.input_blocks.8.0.out_layers.3.bias": "blocks.19.conv2.bias",445 "control_model.input_blocks.8.1.norm.weight": "blocks.20.norm.weight",446 "control_model.input_blocks.8.1.norm.bias": "blocks.20.norm.bias",447 "control_model.input_blocks.8.1.proj_in.weight": "blocks.20.proj_in.weight",448 "control_model.input_blocks.8.1.proj_in.bias": "blocks.20.proj_in.bias",449 "control_model.input_blocks.8.1.transformer_blocks.0.attn1.to_q.weight": "blocks.20.transformer_blocks.0.attn1.to_q.weight",450 "control_model.input_blocks.8.1.transformer_blocks.0.attn1.to_k.weight": "blocks.20.transformer_blocks.0.attn1.to_k.weight",451 "control_model.input_blocks.8.1.transformer_blocks.0.attn1.to_v.weight": "blocks.20.transformer_blocks.0.attn1.to_v.weight",452 "control_model.input_blocks.8.1.transformer_blocks.0.attn1.to_out.0.weight": "blocks.20.transformer_blocks.0.attn1.to_out.weight",453 "control_model.input_blocks.8.1.transformer_blocks.0.attn1.to_out.0.bias": "blocks.20.transformer_blocks.0.attn1.to_out.bias",454 "control_model.input_blocks.8.1.transformer_blocks.0.ff.net.0.proj.weight": "blocks.20.transformer_blocks.0.act_fn.proj.weight",455 "control_model.input_blocks.8.1.transformer_blocks.0.ff.net.0.proj.bias": "blocks.20.transformer_blocks.0.act_fn.proj.bias",456 "control_model.input_blocks.8.1.transformer_blocks.0.ff.net.2.weight": "blocks.20.transformer_blocks.0.ff.weight",457 "control_model.input_blocks.8.1.transformer_blocks.0.ff.net.2.bias": "blocks.20.transformer_blocks.0.ff.bias",458 "control_model.input_blocks.8.1.transformer_blocks.0.attn2.to_q.weight": "blocks.20.transformer_blocks.0.attn2.to_q.weight",459 "control_model.input_blocks.8.1.transformer_blocks.0.attn2.to_k.weight": "blocks.20.transformer_blocks.0.attn2.to_k.weight",460 "control_model.input_blocks.8.1.transformer_blocks.0.attn2.to_v.weight": "blocks.20.transformer_blocks.0.attn2.to_v.weight",461 "control_model.input_blocks.8.1.transformer_blocks.0.attn2.to_out.0.weight": "blocks.20.transformer_blocks.0.attn2.to_out.weight",462 "control_model.input_blocks.8.1.transformer_blocks.0.attn2.to_out.0.bias": "blocks.20.transformer_blocks.0.attn2.to_out.bias",463 "control_model.input_blocks.8.1.transformer_blocks.0.norm1.weight": "blocks.20.transformer_blocks.0.norm1.weight",464 "control_model.input_blocks.8.1.transformer_blocks.0.norm1.bias": "blocks.20.transformer_blocks.0.norm1.bias",465 "control_model.input_blocks.8.1.transformer_blocks.0.norm2.weight": "blocks.20.transformer_blocks.0.norm2.weight",466 "control_model.input_blocks.8.1.transformer_blocks.0.norm2.bias": "blocks.20.transformer_blocks.0.norm2.bias",467 "control_model.input_blocks.8.1.transformer_blocks.0.norm3.weight": "blocks.20.transformer_blocks.0.norm3.weight",468 "control_model.input_blocks.8.1.transformer_blocks.0.norm3.bias": "blocks.20.transformer_blocks.0.norm3.bias",469 "control_model.input_blocks.8.1.proj_out.weight": "blocks.20.proj_out.weight",470 "control_model.input_blocks.8.1.proj_out.bias": "blocks.20.proj_out.bias",471 "control_model.input_blocks.9.0.op.weight": "blocks.22.conv.weight",472 "control_model.input_blocks.9.0.op.bias": "blocks.22.conv.bias",473 "control_model.input_blocks.10.0.in_layers.0.weight": "blocks.24.norm1.weight",474 "control_model.input_blocks.10.0.in_layers.0.bias": "blocks.24.norm1.bias",475 "control_model.input_blocks.10.0.in_layers.2.weight": "blocks.24.conv1.weight",476 "control_model.input_blocks.10.0.in_layers.2.bias": "blocks.24.conv1.bias",477 "control_model.input_blocks.10.0.emb_layers.1.weight": "blocks.24.time_emb_proj.weight",478 "control_model.input_blocks.10.0.emb_layers.1.bias": "blocks.24.time_emb_proj.bias",479 "control_model.input_blocks.10.0.out_layers.0.weight": "blocks.24.norm2.weight",480 "control_model.input_blocks.10.0.out_layers.0.bias": "blocks.24.norm2.bias",481 "control_model.input_blocks.10.0.out_layers.3.weight": "blocks.24.conv2.weight",482 "control_model.input_blocks.10.0.out_layers.3.bias": "blocks.24.conv2.bias",483 "control_model.input_blocks.11.0.in_layers.0.weight": "blocks.26.norm1.weight",484 "control_model.input_blocks.11.0.in_layers.0.bias": "blocks.26.norm1.bias",485 "control_model.input_blocks.11.0.in_layers.2.weight": "blocks.26.conv1.weight",486 "control_model.input_blocks.11.0.in_layers.2.bias": "blocks.26.conv1.bias",487 "control_model.input_blocks.11.0.emb_layers.1.weight": "blocks.26.time_emb_proj.weight",488 "control_model.input_blocks.11.0.emb_layers.1.bias": "blocks.26.time_emb_proj.bias",489 "control_model.input_blocks.11.0.out_layers.0.weight": "blocks.26.norm2.weight",490 "control_model.input_blocks.11.0.out_layers.0.bias": "blocks.26.norm2.bias",491 "control_model.input_blocks.11.0.out_layers.3.weight": "blocks.26.conv2.weight",492 "control_model.input_blocks.11.0.out_layers.3.bias": "blocks.26.conv2.bias",493 "control_model.zero_convs.0.0.weight": "controlnet_blocks.0.weight",494 "control_model.zero_convs.0.0.bias": "controlnet_blocks.0.bias",495 "control_model.zero_convs.1.0.weight": "controlnet_blocks.1.weight",496 "control_model.zero_convs.1.0.bias": "controlnet_blocks.0.bias",497 "control_model.zero_convs.2.0.weight": "controlnet_blocks.2.weight",498 "control_model.zero_convs.2.0.bias": "controlnet_blocks.0.bias",499 "control_model.zero_convs.3.0.weight": "controlnet_blocks.3.weight",500 "control_model.zero_convs.3.0.bias": "controlnet_blocks.0.bias",501 "control_model.zero_convs.4.0.weight": "controlnet_blocks.4.weight",502 "control_model.zero_convs.4.0.bias": "controlnet_blocks.4.bias",503 "control_model.zero_convs.5.0.weight": "controlnet_blocks.5.weight",504 "control_model.zero_convs.5.0.bias": "controlnet_blocks.4.bias",505 "control_model.zero_convs.6.0.weight": "controlnet_blocks.6.weight",506 "control_model.zero_convs.6.0.bias": "controlnet_blocks.4.bias",507 "control_model.zero_convs.7.0.weight": "controlnet_blocks.7.weight",508 "control_model.zero_convs.7.0.bias": "controlnet_blocks.7.bias",509 "control_model.zero_convs.8.0.weight": "controlnet_blocks.8.weight",510 "control_model.zero_convs.8.0.bias": "controlnet_blocks.7.bias",511 "control_model.zero_convs.9.0.weight": "controlnet_blocks.9.weight",512 "control_model.zero_convs.9.0.bias": "controlnet_blocks.7.bias",513 "control_model.zero_convs.10.0.weight": "controlnet_blocks.10.weight",514 "control_model.zero_convs.10.0.bias": "controlnet_blocks.7.bias",515 "control_model.zero_convs.11.0.weight": "controlnet_blocks.11.weight",516 "control_model.zero_convs.11.0.bias": "controlnet_blocks.7.bias",517 "control_model.input_hint_block.0.weight": "controlnet_conv_in.blocks.0.weight",518 "control_model.input_hint_block.0.bias": "controlnet_conv_in.blocks.0.bias",519 "control_model.input_hint_block.2.weight": "controlnet_conv_in.blocks.2.weight",520 "control_model.input_hint_block.2.bias": "controlnet_conv_in.blocks.2.bias",521 "control_model.input_hint_block.4.weight": "controlnet_conv_in.blocks.4.weight",522 "control_model.input_hint_block.4.bias": "controlnet_conv_in.blocks.4.bias",523 "control_model.input_hint_block.6.weight": "controlnet_conv_in.blocks.6.weight",524 "control_model.input_hint_block.6.bias": "controlnet_conv_in.blocks.6.bias",525 "control_model.input_hint_block.8.weight": "controlnet_conv_in.blocks.8.weight",526 "control_model.input_hint_block.8.bias": "controlnet_conv_in.blocks.8.bias",527 "control_model.input_hint_block.10.weight": "controlnet_conv_in.blocks.10.weight",528 "control_model.input_hint_block.10.bias": "controlnet_conv_in.blocks.10.bias",529 "control_model.input_hint_block.12.weight": "controlnet_conv_in.blocks.12.weight",530 "control_model.input_hint_block.12.bias": "controlnet_conv_in.blocks.12.bias",531 "control_model.input_hint_block.14.weight": "controlnet_conv_in.blocks.14.weight",532 "control_model.input_hint_block.14.bias": "controlnet_conv_in.blocks.14.bias",533 "control_model.middle_block.0.in_layers.0.weight": "blocks.28.norm1.weight",534 "control_model.middle_block.0.in_layers.0.bias": "blocks.28.norm1.bias",535 "control_model.middle_block.0.in_layers.2.weight": "blocks.28.conv1.weight",536 "control_model.middle_block.0.in_layers.2.bias": "blocks.28.conv1.bias",537 "control_model.middle_block.0.emb_layers.1.weight": "blocks.28.time_emb_proj.weight",538 "control_model.middle_block.0.emb_layers.1.bias": "blocks.28.time_emb_proj.bias",539 "control_model.middle_block.0.out_layers.0.weight": "blocks.28.norm2.weight",540 "control_model.middle_block.0.out_layers.0.bias": "blocks.28.norm2.bias",541 "control_model.middle_block.0.out_layers.3.weight": "blocks.28.conv2.weight",542 "control_model.middle_block.0.out_layers.3.bias": "blocks.28.conv2.bias",543 "control_model.middle_block.1.norm.weight": "blocks.29.norm.weight",544 "control_model.middle_block.1.norm.bias": "blocks.29.norm.bias",545 "control_model.middle_block.1.proj_in.weight": "blocks.29.proj_in.weight",546 "control_model.middle_block.1.proj_in.bias": "blocks.29.proj_in.bias",547 "control_model.middle_block.1.transformer_blocks.0.attn1.to_q.weight": "blocks.29.transformer_blocks.0.attn1.to_q.weight",548 "control_model.middle_block.1.transformer_blocks.0.attn1.to_k.weight": "blocks.29.transformer_blocks.0.attn1.to_k.weight",549 "control_model.middle_block.1.transformer_blocks.0.attn1.to_v.weight": "blocks.29.transformer_blocks.0.attn1.to_v.weight",550 "control_model.middle_block.1.transformer_blocks.0.attn1.to_out.0.weight": "blocks.29.transformer_blocks.0.attn1.to_out.weight",551 "control_model.middle_block.1.transformer_blocks.0.attn1.to_out.0.bias": "blocks.29.transformer_blocks.0.attn1.to_out.bias",552 "control_model.middle_block.1.transformer_blocks.0.ff.net.0.proj.weight": "blocks.29.transformer_blocks.0.act_fn.proj.weight",553 "control_model.middle_block.1.transformer_blocks.0.ff.net.0.proj.bias": "blocks.29.transformer_blocks.0.act_fn.proj.bias",554 "control_model.middle_block.1.transformer_blocks.0.ff.net.2.weight": "blocks.29.transformer_blocks.0.ff.weight",555 "control_model.middle_block.1.transformer_blocks.0.ff.net.2.bias": "blocks.29.transformer_blocks.0.ff.bias",556 "control_model.middle_block.1.transformer_blocks.0.attn2.to_q.weight": "blocks.29.transformer_blocks.0.attn2.to_q.weight",557 "control_model.middle_block.1.transformer_blocks.0.attn2.to_k.weight": "blocks.29.transformer_blocks.0.attn2.to_k.weight",558 "control_model.middle_block.1.transformer_blocks.0.attn2.to_v.weight": "blocks.29.transformer_blocks.0.attn2.to_v.weight",559 "control_model.middle_block.1.transformer_blocks.0.attn2.to_out.0.weight": "blocks.29.transformer_blocks.0.attn2.to_out.weight",560 "control_model.middle_block.1.transformer_blocks.0.attn2.to_out.0.bias": "blocks.29.transformer_blocks.0.attn2.to_out.bias",561 "control_model.middle_block.1.transformer_blocks.0.norm1.weight": "blocks.29.transformer_blocks.0.norm1.weight",562 "control_model.middle_block.1.transformer_blocks.0.norm1.bias": "blocks.29.transformer_blocks.0.norm1.bias",563 "control_model.middle_block.1.transformer_blocks.0.norm2.weight": "blocks.29.transformer_blocks.0.norm2.weight",564 "control_model.middle_block.1.transformer_blocks.0.norm2.bias": "blocks.29.transformer_blocks.0.norm2.bias",565 "control_model.middle_block.1.transformer_blocks.0.norm3.weight": "blocks.29.transformer_blocks.0.norm3.weight",566 "control_model.middle_block.1.transformer_blocks.0.norm3.bias": "blocks.29.transformer_blocks.0.norm3.bias",567 "control_model.middle_block.1.proj_out.weight": "blocks.29.proj_out.weight",568 "control_model.middle_block.1.proj_out.bias": "blocks.29.proj_out.bias",569 "control_model.middle_block.2.in_layers.0.weight": "blocks.30.norm1.weight",570 "control_model.middle_block.2.in_layers.0.bias": "blocks.30.norm1.bias",571 "control_model.middle_block.2.in_layers.2.weight": "blocks.30.conv1.weight",572 "control_model.middle_block.2.in_layers.2.bias": "blocks.30.conv1.bias",573 "control_model.middle_block.2.emb_layers.1.weight": "blocks.30.time_emb_proj.weight",574 "control_model.middle_block.2.emb_layers.1.bias": "blocks.30.time_emb_proj.bias",575 "control_model.middle_block.2.out_layers.0.weight": "blocks.30.norm2.weight",576 "control_model.middle_block.2.out_layers.0.bias": "blocks.30.norm2.bias",577 "control_model.middle_block.2.out_layers.3.weight": "blocks.30.conv2.weight",578 "control_model.middle_block.2.out_layers.3.bias": "blocks.30.conv2.bias",579 "control_model.middle_block_out.0.weight": "controlnet_blocks.12.weight",580 "control_model.middle_block_out.0.bias": "controlnet_blocks.7.bias",581 }582 state_dict_ = {}583 for name in state_dict:584 if name in rename_dict:585 param = state_dict[name]586 if ".proj_in." in name or ".proj_out." in name:587 param = param.squeeze()588 state_dict_[rename_dict[name]] = param589 return state_dict_590