Make it easier to implement future qwen controlnets. (Comfy-Org#9485)

Author: comfyanonymous

comfy/controlnet.py

@@ -236,11 +236,11 @@ def get_control(self, x_noisy, t, cond, batched_number, transformer_options):
             self.cond_hint = None
             compression_ratio = self.compression_ratio
             if self.vae is not None:
-                compression_ratio *= self.vae.downscale_ratio
+                compression_ratio *= self.vae.spacial_compression_encode()
             else:
                 if self.latent_format is not None:
                     raise ValueError("This Controlnet needs a VAE but none was provided, please use a ControlNetApply node with a VAE input and connect it.")
-            self.cond_hint = comfy.utils.common_upscale(self.cond_hint_original, x_noisy.shape[3] * compression_ratio, x_noisy.shape[2] * compression_ratio, self.upscale_algorithm, "center")
+            self.cond_hint = comfy.utils.common_upscale(self.cond_hint_original, x_noisy.shape[-1] * compression_ratio, x_noisy.shape[-2] * compression_ratio, self.upscale_algorithm, "center")
             self.cond_hint = self.preprocess_image(self.cond_hint)
             if self.vae is not None:
                 loaded_models = comfy.model_management.loaded_models(only_currently_used=True)

comfy/ldm/qwen_image/model.py

@@ -293,13 +293,15 @@ def __init__(
         guidance_embeds: bool = False,
         axes_dims_rope: Tuple[int, int, int] = (16, 56, 56),
         image_model=None,
+        final_layer=True,
         dtype=None,
         device=None,
         operations=None,
     ):
         super().__init__()
         self.dtype = dtype
         self.patch_size = patch_size
+        self.in_channels = in_channels
         self.out_channels = out_channels or in_channels
         self.inner_dim = num_attention_heads * attention_head_dim
 
@@ -329,9 +331,9 @@ def __init__(
             for _ in range(num_layers)
         ])
 
-        self.norm_out = LastLayer(self.inner_dim, self.inner_dim, dtype=dtype, device=device, operations=operations)
-        self.proj_out = operations.Linear(self.inner_dim, patch_size * patch_size * self.out_channels, bias=True, dtype=dtype, device=device)
-        self.gradient_checkpointing = False
+        if final_layer:
+            self.norm_out = LastLayer(self.inner_dim, self.inner_dim, dtype=dtype, device=device, operations=operations)
+            self.proj_out = operations.Linear(self.inner_dim, patch_size * patch_size * self.out_channels, bias=True, dtype=dtype, device=device)
 
     def process_img(self, x, index=0, h_offset=0, w_offset=0):
         bs, c, t, h, w = x.shape
@@ -362,6 +364,7 @@ def forward(
         guidance: torch.Tensor = None,
         ref_latents=None,
         transformer_options={},
+        control=None,
         **kwargs
     ):
         timestep = timesteps
@@ -443,6 +446,13 @@ def block_wrap(args):
                     hidden_states = out["img"]
                     encoder_hidden_states = out["txt"]
 
+            if control is not None: # Controlnet
+                control_i = control.get("input")
+                if i < len(control_i):
+                    add = control_i[i]
+                    if add is not None:
+                        hidden_states += add
+
         hidden_states = self.norm_out(hidden_states, temb)
         hidden_states = self.proj_out(hidden_states)
 

comfy/model_detection.py

@@ -492,6 +492,8 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
     if '{}txt_norm.weight'.format(key_prefix) in state_dict_keys:  # Qwen Image
         dit_config = {}
         dit_config["image_model"] = "qwen_image"
+        dit_config["in_channels"] = state_dict['{}img_in.weight'.format(key_prefix)].shape[1]
+        dit_config["num_layers"] = count_blocks(state_dict_keys, '{}transformer_blocks.'.format(key_prefix) + '{}.')
         return dit_config
 
     if '{}input_blocks.0.0.weight'.format(key_prefix) not in state_dict_keys: