ComfyUI version 0.3.39

* Added initial Flux.1 Kontext Pro Image node - recreated branch to save myself sanity from rebase crap after master got rebased

* Add safety filter to Kontext.

* Make safety = 2 and input image is optional.

* Add BFL kontext API nodes.

---------

Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>

comfy/conds.py

@@ -24,6 +24,10 @@ class CONDRegular:
             conds.append(x.cond)
         return torch.cat(conds)
 
+    def size(self):
+        return list(self.cond.size())
+
+
 class CONDNoiseShape(CONDRegular):
     def process_cond(self, batch_size, device, area, **kwargs):
         data = self.cond
@@ -64,6 +68,7 @@ class CONDCrossAttn(CONDRegular):
             out.append(c)
         return torch.cat(out)
 
+
 class CONDConstant(CONDRegular):
     def __init__(self, cond):
         self.cond = cond
@@ -78,3 +83,6 @@ class CONDConstant(CONDRegular):
 
     def concat(self, others):
         return self.cond
+
+    def size(self):
+        return [1]

comfy/ldm/wan/model.py

@@ -539,13 +539,20 @@ class WanModel(torch.nn.Module):
         x = self.unpatchify(x, grid_sizes)
         return x
 
-    def forward(self, x, timestep, context, clip_fea=None, transformer_options={}, **kwargs):
+    def forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, transformer_options={}, **kwargs):
         bs, c, t, h, w = x.shape
         x = comfy.ldm.common_dit.pad_to_patch_size(x, self.patch_size)
+
         patch_size = self.patch_size
         t_len = ((t + (patch_size[0] // 2)) // patch_size[0])
         h_len = ((h + (patch_size[1] // 2)) // patch_size[1])
         w_len = ((w + (patch_size[2] // 2)) // patch_size[2])
+
+        if time_dim_concat is not None:
+            time_dim_concat = comfy.ldm.common_dit.pad_to_patch_size(time_dim_concat, self.patch_size)
+            x = torch.cat([x, time_dim_concat], dim=2)
+            t_len = ((x.shape[2] + (patch_size[0] // 2)) // patch_size[0])
+
         img_ids = torch.zeros((t_len, h_len, w_len, 3), device=x.device, dtype=x.dtype)
         img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + torch.linspace(0, t_len - 1, steps=t_len, device=x.device, dtype=x.dtype).reshape(-1, 1, 1)
         img_ids[:, :, :, 1] = img_ids[:, :, :, 1] + torch.linspace(0, h_len - 1, steps=h_len, device=x.device, dtype=x.dtype).reshape(1, -1, 1)

comfy/lora.py

@@ -283,8 +283,9 @@ def model_lora_keys_unet(model, key_map={}):
         for k in sdk:
             if k.startswith("diffusion_model."):
                 if k.endswith(".weight"):
-                    key_lora = k[len("diffusion_model."):-len(".weight")].replace(".", "_")
+                    key_lora = k[len("diffusion_model."):-len(".weight")]
-                    key_map["lycoris_{}".format(key_lora)] = k #SimpleTuner lycoris format
+                    key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = k #SimpleTuner lycoris format
+                    key_map["transformer.{}".format(key_lora)] = k #SimpleTuner regular format
 
     if isinstance(model, comfy.model_base.ACEStep):
         for k in sdk:

comfy/model_base.py

@@ -135,6 +135,7 @@ class BaseModel(torch.nn.Module):
         logging.info("model_type {}".format(model_type.name))
         logging.debug("adm {}".format(self.adm_channels))
         self.memory_usage_factor = model_config.memory_usage_factor
+        self.memory_usage_factor_conds = ()
 
     def apply_model(self, x, t, c_concat=None, c_crossattn=None, control=None, transformer_options={}, **kwargs):
         return comfy.patcher_extension.WrapperExecutor.new_class_executor(
@@ -325,19 +326,28 @@ class BaseModel(torch.nn.Module):
     def scale_latent_inpaint(self, sigma, noise, latent_image, **kwargs):
         return self.model_sampling.noise_scaling(sigma.reshape([sigma.shape[0]] + [1] * (len(noise.shape) - 1)), noise, latent_image)
 
-    def memory_required(self, input_shape):
+    def memory_required(self, input_shape, cond_shapes={}):
+        input_shapes = [input_shape]
+        for c in self.memory_usage_factor_conds:
+            shape = cond_shapes.get(c, None)
+            if shape is not None and len(shape) > 0:
+                input_shapes += shape
+
         if comfy.model_management.xformers_enabled() or comfy.model_management.pytorch_attention_flash_attention():
             dtype = self.get_dtype()
             if self.manual_cast_dtype is not None:
                 dtype = self.manual_cast_dtype
             #TODO: this needs to be tweaked
-            area = input_shape[0] * math.prod(input_shape[2:])
+            area = sum(map(lambda input_shape: input_shape[0] * math.prod(input_shape[2:]), input_shapes))
             return (area * comfy.model_management.dtype_size(dtype) * 0.01 * self.memory_usage_factor) * (1024 * 1024)
         else:
             #TODO: this formula might be too aggressive since I tweaked the sub-quad and split algorithms to use less memory.
-            area = input_shape[0] * math.prod(input_shape[2:])
+            area = sum(map(lambda input_shape: input_shape[0] * math.prod(input_shape[2:]), input_shapes))
             return (area * 0.15 * self.memory_usage_factor) * (1024 * 1024)
 
+    def extra_conds_shapes(self, **kwargs):
+        return {}
+
 
 def unclip_adm(unclip_conditioning, device, noise_augmentor, noise_augment_merge=0.0, seed=None):
     adm_inputs = []
@@ -1047,6 +1057,11 @@ class WAN21(BaseModel):
         clip_vision_output = kwargs.get("clip_vision_output", None)
         if clip_vision_output is not None:
             out['clip_fea'] = comfy.conds.CONDRegular(clip_vision_output.penultimate_hidden_states)
+
+        time_dim_concat = kwargs.get("time_dim_concat", None)
+        if time_dim_concat is not None:
+            out['time_dim_concat'] = comfy.conds.CONDRegular(self.process_latent_in(time_dim_concat))
+
         return out
 
 

comfy/sampler_helpers.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
 import uuid
+import math
+import collections
 import comfy.model_management
 import comfy.conds
 import comfy.utils
@@ -104,6 +106,21 @@ def cleanup_additional_models(models):
         if hasattr(m, 'cleanup'):
             m.cleanup()
 
+def estimate_memory(model, noise_shape, conds):
+    cond_shapes = collections.defaultdict(list)
+    cond_shapes_min = {}
+    for _, cs in conds.items():
+        for cond in cs:
+            for k, v in model.model.extra_conds_shapes(**cond).items():
+                cond_shapes[k].append(v)
+                if cond_shapes_min.get(k, None) is None:
+                    cond_shapes_min[k] = [v]
+                elif math.prod(v) > math.prod(cond_shapes_min[k][0]):
+                    cond_shapes_min[k] = [v]
+
+    memory_required = model.model.memory_required([noise_shape[0] * 2] + list(noise_shape[1:]), cond_shapes=cond_shapes)
+    minimum_memory_required = model.model.memory_required([noise_shape[0]] + list(noise_shape[1:]), cond_shapes=cond_shapes_min)
+    return memory_required, minimum_memory_required
 
 def prepare_sampling(model: ModelPatcher, noise_shape, conds, model_options=None):
     executor = comfy.patcher_extension.WrapperExecutor.new_executor(
@@ -117,9 +134,8 @@ def _prepare_sampling(model: ModelPatcher, noise_shape, conds, model_options=Non
     models, inference_memory = get_additional_models(conds, model.model_dtype())
     models += get_additional_models_from_model_options(model_options)
     models += model.get_nested_additional_models()  # TODO: does this require inference_memory update?
-    memory_required = model.memory_required([noise_shape[0] * 2] + list(noise_shape[1:])) + inference_memory
+    memory_required, minimum_memory_required = estimate_memory(model, noise_shape, conds)
-    minimum_memory_required = model.memory_required([noise_shape[0]] + list(noise_shape[1:])) + inference_memory
+    comfy.model_management.load_models_gpu([model] + models, memory_required=memory_required + inference_memory, minimum_memory_required=minimum_memory_required + inference_memory)
-    comfy.model_management.load_models_gpu([model] + models, memory_required=memory_required, minimum_memory_required=minimum_memory_required)
     real_model = model.model
 
     return real_model, conds, models

comfy/samplers.py

@@ -256,7 +256,13 @@ def _calc_cond_batch(model: 'BaseModel', conds: list[list[dict]], x_in: torch.Te
             for i in range(1, len(to_batch_temp) + 1):
                 batch_amount = to_batch_temp[:len(to_batch_temp)//i]
                 input_shape = [len(batch_amount) * first_shape[0]] + list(first_shape)[1:]
-                if model.memory_required(input_shape) * 1.5 < free_memory:
+                cond_shapes = collections.defaultdict(list)
+                for tt in batch_amount:
+                    cond = {k: v.size() for k, v in to_run[tt][0].conditioning.items()}
+                    for k, v in to_run[tt][0].conditioning.items():
+                        cond_shapes[k].append(v.size())
+
+                if model.memory_required(input_shape, cond_shapes=cond_shapes) * 1.5 < free_memory:
                     to_batch = batch_amount
                     break
 

comfy/text_encoders/long_clipl.json

@@ -1,25 +0,0 @@
-{
-  "_name_or_path": "openai/clip-vit-large-patch14",
-  "architectures": [
-    "CLIPTextModel"
-  ],
-  "attention_dropout": 0.0,
-  "bos_token_id": 0,
-  "dropout": 0.0,
-  "eos_token_id": 49407,
-  "hidden_act": "quick_gelu",
-  "hidden_size": 768,
-  "initializer_factor": 1.0,
-  "initializer_range": 0.02,
-  "intermediate_size": 3072,
-  "layer_norm_eps": 1e-05,
-  "max_position_embeddings": 248,
-  "model_type": "clip_text_model",
-  "num_attention_heads": 12,
-  "num_hidden_layers": 12,
-  "pad_token_id": 1,
-  "projection_dim": 768,
-  "torch_dtype": "float32",
-  "transformers_version": "4.24.0",
-  "vocab_size": 49408
-}

comfy_api_nodes/apis/bfl_api.py

@@ -108,6 +108,40 @@ class BFLFluxProGenerateRequest(BaseModel):
     # )
 
 
+class BFLFluxKontextProGenerateRequest(BaseModel):
+    prompt: str = Field(..., description='The text prompt for what you wannt to edit.')
+    input_image: Optional[str] = Field(None, description='Image to edit in base64 format')
+    seed: Optional[int] = Field(None, description='The seed value for reproducibility.')
+    guidance: confloat(ge=0.1, le=99.0) = Field(..., description='Guidance strength for the image generation process')
+    steps: conint(ge=1, le=150) = Field(..., description='Number of steps for the image generation process')
+    safety_tolerance: Optional[conint(ge=0, le=2)] = Field(
+        2, description='Tolerance level for input and output moderation. Between 0 and 2, 0 being most strict, 6 being least strict. Defaults to 2.'
+    )
+    output_format: Optional[BFLOutputFormat] = Field(
+        BFLOutputFormat.png, description="Output format for the generated image. Can be 'jpeg' or 'png'.", examples=['png']
+    )
+    aspect_ratio: Optional[str] = Field(None, description='Aspect ratio of the image between 21:9 and 9:21.')
+    prompt_upsampling: Optional[bool] = Field(
+        None, description='Whether to perform upsampling on the prompt. If active, automatically modifies the prompt for more creative generation.'
+    )
+
+class BFLFluxKontextMaxGenerateRequest(BaseModel):
+    prompt: str = Field(..., description='The text prompt for what you wannt to edit.')
+    input_image: Optional[str] = Field(None, description='Image to edit in base64 format')
+    seed: Optional[int] = Field(None, description='The seed value for reproducibility.')
+    guidance: confloat(ge=0.1, le=99.0) = Field(..., description='Guidance strength for the image generation process')
+    steps: conint(ge=1, le=150) = Field(..., description='Number of steps for the image generation process')
+    safety_tolerance: Optional[conint(ge=0, le=2)] = Field(
+        2, description='Tolerance level for input and output moderation. Between 0 and 2, 0 being most strict, 6 being least strict. Defaults to 2.'
+    )
+    output_format: Optional[BFLOutputFormat] = Field(
+        BFLOutputFormat.png, description="Output format for the generated image. Can be 'jpeg' or 'png'.", examples=['png']
+    )
+    aspect_ratio: Optional[str] = Field(None, description='Aspect ratio of the image between 21:9 and 9:21.')
+    prompt_upsampling: Optional[bool] = Field(
+        None, description='Whether to perform upsampling on the prompt. If active, automatically modifies the prompt for more creative generation.'
+    )
+
 class BFLFluxProUltraGenerateRequest(BaseModel):
     prompt: str = Field(..., description='The text prompt for image generation.')
     prompt_upsampling: Optional[bool] = Field(

comfy_api_nodes/nodes_bfl.py

@@ -1,6 +1,6 @@
 import io
 from inspect import cleandoc
-from typing import Union
+from typing import Union, Optional
 from comfy.comfy_types.node_typing import IO, ComfyNodeABC
 from comfy_api_nodes.apis.bfl_api import (
     BFLStatus,
@@ -9,6 +9,7 @@ from comfy_api_nodes.apis.bfl_api import (
     BFLFluxCannyImageRequest,
     BFLFluxDepthImageRequest,
     BFLFluxProGenerateRequest,
+    BFLFluxKontextProGenerateRequest,
     BFLFluxProUltraGenerateRequest,
     BFLFluxProGenerateResponse,
 )
@@ -269,6 +270,287 @@ class FluxProUltraImageNode(ComfyNodeABC):
         return (output_image,)
 
 
+class FluxKontextProImageNode(ComfyNodeABC):
+    """
+    Edits images using Flux.1 Kontext Pro via api based on prompt and resolution.
+    """
+
+    MINIMUM_RATIO = 1 / 4
+    MAXIMUM_RATIO = 4 / 1
+    MINIMUM_RATIO_STR = "1:4"
+    MAXIMUM_RATIO_STR = "4:1"
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "prompt": (
+                    IO.STRING,
+                    {
+                        "multiline": True,
+                        "default": "",
+                        "tooltip": "Prompt for the image generation - specify what and how to edit.",
+                    },
+                ),
+                "aspect_ratio": (
+                    IO.STRING,
+                    {
+                        "default": "16:9",
+                        "tooltip": "Aspect ratio of image; must be between 1:4 and 4:1.",
+                    },
+                ),
+                "guidance": (
+                    IO.FLOAT,
+                    {
+                        "default": 3.0,
+                        "min": 0.1,
+                        "max": 99.0,
+                        "step": 0.1,
+                        "tooltip": "Guidance strength for the image generation process"
+                    },
+                ),
+                "steps": (
+                    IO.INT,
+                    {
+                        "default": 50,
+                        "min": 1,
+                        "max": 150,
+                        "tooltip": "Number of steps for the image generation process"
+                    },
+                ),
+                "seed": (
+                    IO.INT,
+                    {
+                        "default": 0,
+                        "min": 0,
+                        "max": 0xFFFFFFFFFFFFFFFF,
+                        "control_after_generate": True,
+                        "tooltip": "The random seed used for creating the noise.",
+                    },
+                ),
+                "prompt_upsampling": (
+                    IO.BOOLEAN,
+                    {
+                        "default": False,
+                        "tooltip": "Whether to perform upsampling on the prompt. If active, automatically modifies the prompt for more creative generation, but results are nondeterministic (same seed will not produce exactly the same result).",
+                    },
+                ),
+            },
+            "optional": {
+                "input_image": (IO.IMAGE,),
+            },
+            "hidden": {
+                "auth_token": "AUTH_TOKEN_COMFY_ORG",
+                "comfy_api_key": "API_KEY_COMFY_ORG",
+                "unique_id": "UNIQUE_ID",
+            },
+        }
+
+    @classmethod
+    def VALIDATE_INPUTS(cls, aspect_ratio: str):
+        try:
+            validate_aspect_ratio(
+                aspect_ratio,
+                minimum_ratio=cls.MINIMUM_RATIO,
+                maximum_ratio=cls.MAXIMUM_RATIO,
+                minimum_ratio_str=cls.MINIMUM_RATIO_STR,
+                maximum_ratio_str=cls.MAXIMUM_RATIO_STR,
+            )
+        except Exception as e:
+            return str(e)
+        return True
+
+    RETURN_TYPES = (IO.IMAGE,)
+    DESCRIPTION = cleandoc(__doc__ or "")  # Handle potential None value
+    FUNCTION = "api_call"
+    API_NODE = True
+    CATEGORY = "api node/image/BFL"
+
+    def api_call(
+        self,
+        prompt: str,
+        aspect_ratio: str,
+        guidance: float,
+        steps: int,
+        input_image: Optional[torch.Tensor]=None,
+        seed=0,
+        prompt_upsampling=False,
+        unique_id: Union[str, None] = None,
+        **kwargs,
+    ):
+        if input_image is None:
+            validate_string(prompt, strip_whitespace=False)
+        operation = SynchronousOperation(
+            endpoint=ApiEndpoint(
+                path="/proxy/bfl/flux-kontext-pro/generate",
+                method=HttpMethod.POST,
+                request_model=BFLFluxKontextProGenerateRequest,
+                response_model=BFLFluxProGenerateResponse,
+            ),
+            request=BFLFluxKontextProGenerateRequest(
+                prompt=prompt,
+                prompt_upsampling=prompt_upsampling,
+                guidance=round(guidance, 1),
+                steps=steps,
+                seed=seed,
+                aspect_ratio=validate_aspect_ratio(
+                    aspect_ratio,
+                    minimum_ratio=self.MINIMUM_RATIO,
+                    maximum_ratio=self.MAXIMUM_RATIO,
+                    minimum_ratio_str=self.MINIMUM_RATIO_STR,
+                    maximum_ratio_str=self.MAXIMUM_RATIO_STR,
+                ),
+                input_image=(
+                    input_image
+                    if input_image is None
+                    else convert_image_to_base64(input_image)
+                )
+            ),
+            auth_kwargs=kwargs,
+        )
+        output_image = handle_bfl_synchronous_operation(operation, node_id=unique_id)
+        return (output_image,)
+
+class FluxKontextMaxImageNode(ComfyNodeABC):
+    """
+    Edits images using Flux.1 Kontext Max via api based on prompt and resolution.
+    """
+
+    MINIMUM_RATIO = 1 / 4
+    MAXIMUM_RATIO = 4 / 1
+    MINIMUM_RATIO_STR = "1:4"
+    MAXIMUM_RATIO_STR = "4:1"
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "prompt": (
+                    IO.STRING,
+                    {
+                        "multiline": True,
+                        "default": "",
+                        "tooltip": "Prompt for the image generation - specify what and how to edit.",
+                    },
+                ),
+                "aspect_ratio": (
+                    IO.STRING,
+                    {
+                        "default": "16:9",
+                        "tooltip": "Aspect ratio of image; must be between 1:4 and 4:1.",
+                    },
+                ),
+                "guidance": (
+                    IO.FLOAT,
+                    {
+                        "default": 3.0,
+                        "min": 0.1,
+                        "max": 99.0,
+                        "step": 0.1,
+                        "tooltip": "Guidance strength for the image generation process"
+                    },
+                ),
+                "steps": (
+                    IO.INT,
+                    {
+                        "default": 50,
+                        "min": 1,
+                        "max": 150,
+                        "tooltip": "Number of steps for the image generation process"
+                    },
+                ),
+                "seed": (
+                    IO.INT,
+                    {
+                        "default": 0,
+                        "min": 0,
+                        "max": 0xFFFFFFFFFFFFFFFF,
+                        "control_after_generate": True,
+                        "tooltip": "The random seed used for creating the noise.",
+                    },
+                ),
+                "prompt_upsampling": (
+                    IO.BOOLEAN,
+                    {
+                        "default": False,
+                        "tooltip": "Whether to perform upsampling on the prompt. If active, automatically modifies the prompt for more creative generation, but results are nondeterministic (same seed will not produce exactly the same result).",
+                    },
+                ),
+            },
+            "optional": {
+                "input_image": (IO.IMAGE,),
+            },
+            "hidden": {
+                "auth_token": "AUTH_TOKEN_COMFY_ORG",
+                "comfy_api_key": "API_KEY_COMFY_ORG",
+                "unique_id": "UNIQUE_ID",
+            },
+        }
+
+    @classmethod
+    def VALIDATE_INPUTS(cls, aspect_ratio: str):
+        try:
+            validate_aspect_ratio(
+                aspect_ratio,
+                minimum_ratio=cls.MINIMUM_RATIO,
+                maximum_ratio=cls.MAXIMUM_RATIO,
+                minimum_ratio_str=cls.MINIMUM_RATIO_STR,
+                maximum_ratio_str=cls.MAXIMUM_RATIO_STR,
+            )
+        except Exception as e:
+            return str(e)
+        return True
+
+    RETURN_TYPES = (IO.IMAGE,)
+    DESCRIPTION = cleandoc(__doc__ or "")  # Handle potential None value
+    FUNCTION = "api_call"
+    API_NODE = True
+    CATEGORY = "api node/image/BFL"
+
+    def api_call(
+        self,
+        prompt: str,
+        aspect_ratio: str,
+        guidance: float,
+        steps: int,
+        input_image: Optional[torch.Tensor]=None,
+        seed=0,
+        prompt_upsampling=False,
+        unique_id: Union[str, None] = None,
+        **kwargs,
+    ):
+        if input_image is None:
+            validate_string(prompt, strip_whitespace=False)
+        operation = SynchronousOperation(
+            endpoint=ApiEndpoint(
+                path="/proxy/bfl/flux-kontext-max/generate",
+                method=HttpMethod.POST,
+                request_model=BFLFluxKontextProGenerateRequest,
+                response_model=BFLFluxProGenerateResponse,
+            ),
+            request=BFLFluxKontextProGenerateRequest(
+                prompt=prompt,
+                prompt_upsampling=prompt_upsampling,
+                guidance=round(guidance, 1),
+                steps=steps,
+                seed=seed,
+                aspect_ratio=validate_aspect_ratio(
+                    aspect_ratio,
+                    minimum_ratio=self.MINIMUM_RATIO,
+                    maximum_ratio=self.MAXIMUM_RATIO,
+                    minimum_ratio_str=self.MINIMUM_RATIO_STR,
+                    maximum_ratio_str=self.MAXIMUM_RATIO_STR,
+                ),
+                input_image=(
+                    input_image
+                    if input_image is None
+                    else convert_image_to_base64(input_image)
+                )
+            ),
+            auth_kwargs=kwargs,
+        )
+        output_image = handle_bfl_synchronous_operation(operation, node_id=unique_id)
+        return (output_image,)
 
 class FluxProImageNode(ComfyNodeABC):
     """
@@ -914,6 +1196,8 @@ class FluxProDepthNode(ComfyNodeABC):
 NODE_CLASS_MAPPINGS = {
     "FluxProUltraImageNode": FluxProUltraImageNode,
     # "FluxProImageNode": FluxProImageNode,
+    "FluxKontextProImageNode": FluxKontextProImageNode,
+    "FluxKontextMaxImageNode": FluxKontextMaxImageNode,
     "FluxProExpandNode": FluxProExpandNode,
     "FluxProFillNode": FluxProFillNode,
     "FluxProCannyNode": FluxProCannyNode,
@@ -924,6 +1208,8 @@ NODE_CLASS_MAPPINGS = {
 NODE_DISPLAY_NAME_MAPPINGS = {
     "FluxProUltraImageNode": "Flux 1.1 [pro] Ultra Image",
     # "FluxProImageNode": "Flux 1.1 [pro] Image",
+    "FluxKontextProImageNode": "Flux.1 Kontext Pro Image",
+    "FluxKontextMaxImageNode": "Flux.1 Kontext Max Image",
     "FluxProExpandNode": "Flux.1 Expand Image",
     "FluxProFillNode": "Flux.1 Fill Image",
     "FluxProCannyNode": "Flux.1 Canny Control Image",

comfy_api_nodes/nodes_pika.py

@@ -6,40 +6,42 @@ Pika API docs: https://pika-827374fb.mintlify.app/api-reference
 from __future__ import annotations
 
 import io
-from typing import Optional, TypeVar
 import logging
-import torch
+from typing import Optional, TypeVar
+
 import numpy as np
+import torch
+
+from comfy.comfy_types.node_typing import IO, ComfyNodeABC, InputTypeOptions
+from comfy_api.input_impl import VideoFromFile
+from comfy_api.input_impl.video_types import VideoCodec, VideoContainer, VideoInput
+from comfy_api_nodes.apinode_utils import (
+    download_url_to_video_output,
+    tensor_to_bytesio,
+)
 from comfy_api_nodes.apis import (
-    PikaBodyGenerate22T2vGenerate22T2vPost,
-    PikaGenerateResponse,
-    PikaBodyGenerate22I2vGenerate22I2vPost,
-    PikaVideoResponse,
-    PikaBodyGenerate22C2vGenerate22PikascenesPost,
     IngredientsMode,
-    PikaDurationEnum,
+    PikaBodyGenerate22C2vGenerate22PikascenesPost,
-    PikaResolutionEnum,
+    PikaBodyGenerate22I2vGenerate22I2vPost,
-    PikaBodyGeneratePikaffectsGeneratePikaffectsPost,
-    PikaBodyGeneratePikadditionsGeneratePikadditionsPost,
-    PikaBodyGeneratePikaswapsGeneratePikaswapsPost,
     PikaBodyGenerate22KeyframeGenerate22PikaframesPost,
+    PikaBodyGenerate22T2vGenerate22T2vPost,
+    PikaBodyGeneratePikadditionsGeneratePikadditionsPost,
+    PikaBodyGeneratePikaffectsGeneratePikaffectsPost,
+    PikaBodyGeneratePikaswapsGeneratePikaswapsPost,
+    PikaDurationEnum,
     Pikaffect,
+    PikaGenerateResponse,
+    PikaResolutionEnum,
+    PikaVideoResponse,
 )
 from comfy_api_nodes.apis.client import (
     ApiEndpoint,
-    HttpMethod,
-    SynchronousOperation,
-    PollingOperation,
     EmptyRequest,
-)
+    HttpMethod,
-from comfy_api_nodes.apinode_utils import (
+    PollingOperation,
-    tensor_to_bytesio,
+    SynchronousOperation,
-    download_url_to_video_output,
 )
 from comfy_api_nodes.mapper_utils import model_field_to_node_input
-from comfy_api.input_impl.video_types import VideoInput, VideoContainer, VideoCodec
-from comfy_api.input_impl import VideoFromFile
-from comfy.comfy_types.node_typing import IO, ComfyNodeABC, InputTypeOptions
 
 R = TypeVar("R")
 
@@ -204,6 +206,7 @@ class PikaImageToVideoV2_2(PikaNodeBase):
             "hidden": {
                 "auth_token": "AUTH_TOKEN_COMFY_ORG",
                 "comfy_api_key": "API_KEY_COMFY_ORG",
+                "unique_id": "UNIQUE_ID",
             },
         }
 
@@ -457,7 +460,7 @@ class PikAdditionsNode(PikaNodeBase):
             },
         }
 
-    DESCRIPTION = "Add any object or image into your video. Upload a video and specify what you’d like to add to create a seamlessly integrated result."
+    DESCRIPTION = "Add any object or image into your video. Upload a video and specify what you'd like to add to create a seamlessly integrated result."
 
     def api_call(
         self,

comfy_extras/nodes_wan.py

@@ -345,6 +345,44 @@ class WanCameraImageToVideo:
         out_latent["samples"] = latent
         return (positive, negative, out_latent)
 
+class WanPhantomSubjectToVideo:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {"positive": ("CONDITIONING", ),
+                             "negative": ("CONDITIONING", ),
+                             "vae": ("VAE", ),
+                             "width": ("INT", {"default": 832, "min": 16, "max": nodes.MAX_RESOLUTION, "step": 16}),
+                             "height": ("INT", {"default": 480, "min": 16, "max": nodes.MAX_RESOLUTION, "step": 16}),
+                             "length": ("INT", {"default": 81, "min": 1, "max": nodes.MAX_RESOLUTION, "step": 4}),
+                             "batch_size": ("INT", {"default": 1, "min": 1, "max": 4096}),
+                },
+                "optional": {"images": ("IMAGE", ),
+                }}
+
+    RETURN_TYPES = ("CONDITIONING", "CONDITIONING", "CONDITIONING", "LATENT")
+    RETURN_NAMES = ("positive", "negative_text", "negative_img_text", "latent")
+    FUNCTION = "encode"
+
+    CATEGORY = "conditioning/video_models"
+
+    def encode(self, positive, negative, vae, width, height, length, batch_size, images):
+        latent = torch.zeros([batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8], device=comfy.model_management.intermediate_device())
+        cond2 = negative
+        if images is not None:
+            images = comfy.utils.common_upscale(images[:length].movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
+            latent_images = []
+            for i in images:
+                latent_images += [vae.encode(i.unsqueeze(0)[:, :, :, :3])]
+            concat_latent_image = torch.cat(latent_images, dim=2)
+
+            positive = node_helpers.conditioning_set_values(positive, {"time_dim_concat": concat_latent_image})
+            cond2 = node_helpers.conditioning_set_values(negative, {"time_dim_concat": concat_latent_image})
+            negative = node_helpers.conditioning_set_values(negative, {"time_dim_concat": comfy.latent_formats.Wan21().process_out(torch.zeros_like(concat_latent_image))})
+
+        out_latent = {}
+        out_latent["samples"] = latent
+        return (positive, cond2, negative, out_latent)
+
 NODE_CLASS_MAPPINGS = {
     "WanImageToVideo": WanImageToVideo,
     "WanFunControlToVideo": WanFunControlToVideo,
@@ -353,4 +391,5 @@ NODE_CLASS_MAPPINGS = {
     "WanVaceToVideo": WanVaceToVideo,
     "TrimVideoLatent": TrimVideoLatent,
     "WanCameraImageToVideo": WanCameraImageToVideo,
+    "WanPhantomSubjectToVideo": WanPhantomSubjectToVideo,
 }

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.3.37"
+__version__ = "0.3.39"

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.3.37"
+version = "0.3.39"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.9"

requirements.txt

@@ -1,5 +1,5 @@
-comfyui-frontend-package==1.20.6
+comfyui-frontend-package==1.20.7
-comfyui-workflow-templates==0.1.20
+comfyui-workflow-templates==0.1.22
 torch
 torchsde
 torchvision