Implement wan2.2 camera model. (Comfy-Org#9357)

Use the old WanCameraImageToVideo node.

Author: comfyanonymous

comfy/ldm/wan/model.py

@@ -768,7 +768,12 @@ def __init__(self,
                  operations=None,
                  ):
 
-        super().__init__(model_type='i2v', patch_size=patch_size, text_len=text_len, in_dim=in_dim, dim=dim, ffn_dim=ffn_dim, freq_dim=freq_dim, text_dim=text_dim, out_dim=out_dim, num_heads=num_heads, num_layers=num_layers, window_size=window_size, qk_norm=qk_norm, cross_attn_norm=cross_attn_norm, eps=eps, flf_pos_embed_token_number=flf_pos_embed_token_number, image_model=image_model, device=device, dtype=dtype, operations=operations)
+        if model_type == 'camera':
+            model_type = 'i2v'
+        else:
+            model_type = 't2v'
+
+        super().__init__(model_type=model_type, patch_size=patch_size, text_len=text_len, in_dim=in_dim, dim=dim, ffn_dim=ffn_dim, freq_dim=freq_dim, text_dim=text_dim, out_dim=out_dim, num_heads=num_heads, num_layers=num_layers, window_size=window_size, qk_norm=qk_norm, cross_attn_norm=cross_attn_norm, eps=eps, flf_pos_embed_token_number=flf_pos_embed_token_number, image_model=image_model, device=device, dtype=dtype, operations=operations)
         operation_settings = {"operations": operations, "device": device, "dtype": dtype}
 
         self.control_adapter = WanCamAdapter(in_dim_control_adapter, dim, kernel_size=patch_size[1:], stride=patch_size[1:], operation_settings=operation_settings)

comfy/model_detection.py

@@ -364,7 +364,10 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
             dit_config["vace_in_dim"] = state_dict['{}vace_patch_embedding.weight'.format(key_prefix)].shape[1]
             dit_config["vace_layers"] = count_blocks(state_dict_keys, '{}vace_blocks.'.format(key_prefix) + '{}.')
         elif '{}control_adapter.conv.weight'.format(key_prefix) in state_dict_keys:
-            dit_config["model_type"] = "camera"
+            if '{}img_emb.proj.0.bias'.format(key_prefix) in state_dict_keys:
+                dit_config["model_type"] = "camera"
+            else:
+                dit_config["model_type"] = "camera_2.2"
         else:
             if '{}img_emb.proj.0.bias'.format(key_prefix) in state_dict_keys:
                 dit_config["model_type"] = "i2v"

comfy/supported_models.py

@@ -1046,6 +1046,18 @@ class WAN21_Camera(WAN21_T2V):
     def get_model(self, state_dict, prefix="", device=None):
         out = model_base.WAN21_Camera(self, image_to_video=False, device=device)
         return out
+
+class WAN22_Camera(WAN21_T2V):
+    unet_config = {
+        "image_model": "wan2.1",
+        "model_type": "camera_2.2",
+        "in_dim": 36,
+    }
+
+    def get_model(self, state_dict, prefix="", device=None):
+        out = model_base.WAN21_Camera(self, image_to_video=False, device=device)
+        return out
+
 class WAN21_Vace(WAN21_T2V):
     unet_config = {
         "image_model": "wan2.1",
@@ -1260,6 +1272,6 @@ def clip_target(self, state_dict={}):
         return supported_models_base.ClipTarget(comfy.text_encoders.qwen_image.QwenImageTokenizer, comfy.text_encoders.qwen_image.te(**hunyuan_detect))
 
 
-models = [LotusD, Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, FluxSchnell, GenmoMochi, LTXV, HunyuanVideoSkyreelsI2V, HunyuanVideoI2V, HunyuanVideo, CosmosT2V, CosmosI2V, CosmosT2IPredict2, CosmosI2VPredict2, Lumina2, WAN22_T2V, WAN21_T2V, WAN21_I2V, WAN21_FunControl2V, WAN21_Vace, WAN21_Camera, Hunyuan3Dv2mini, Hunyuan3Dv2, HiDream, Chroma, ACEStep, Omnigen2, QwenImage]
+models = [LotusD, Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, FluxSchnell, GenmoMochi, LTXV, HunyuanVideoSkyreelsI2V, HunyuanVideoI2V, HunyuanVideo, CosmosT2V, CosmosI2V, CosmosT2IPredict2, CosmosI2VPredict2, Lumina2, WAN22_T2V, WAN21_T2V, WAN21_I2V, WAN21_FunControl2V, WAN21_Vace, WAN21_Camera, WAN22_Camera, Hunyuan3Dv2mini, Hunyuan3Dv2, HiDream, Chroma, ACEStep, Omnigen2, QwenImage]
 
 models += [SVD_img2vid]

comfy_extras/nodes_wan.py

@@ -422,9 +422,12 @@ def execute(cls, positive, negative, vae, width, height, length, batch_size, sta
             start_image = comfy.utils.common_upscale(start_image[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
             concat_latent_image = vae.encode(start_image[:, :, :, :3])
             concat_latent[:,:,:concat_latent_image.shape[2]] = concat_latent_image[:,:,:concat_latent.shape[2]]
+            mask = torch.ones((1, 1, latent.shape[2] * 4, latent.shape[-2], latent.shape[-1]))
+            mask[:, :, :start_image.shape[0] + 3] = 0.0
+            mask = mask.view(1, mask.shape[2] // 4, 4, mask.shape[3], mask.shape[4]).transpose(1, 2)
 
-            positive = node_helpers.conditioning_set_values(positive, {"concat_latent_image": concat_latent})
-            negative = node_helpers.conditioning_set_values(negative, {"concat_latent_image": concat_latent})
+            positive = node_helpers.conditioning_set_values(positive, {"concat_latent_image": concat_latent, "concat_mask": mask})
+            negative = node_helpers.conditioning_set_values(negative, {"concat_latent_image": concat_latent, "concat_mask": mask})
 
         if camera_conditions is not None:
             positive = node_helpers.conditioning_set_values(positive, {'camera_conditions': camera_conditions})