import torch from .blend_modes.index import blend_functions from .blend_modes_enum import BlendModes class ImageBlender: def __init__(self): self.blend_functions = blend_functions @classmethod def INPUT_TYPES(cls): return { "required": { "base_image": ("IMAGE",), "blend_image": ("IMAGE",), "strength": ("FLOAT", { "default": 1, "min": 0.0, "max": 1.0, "step": 0.01 }), "blend_mode": ( [mode.value for mode in BlendModes], {"default": BlendModes.MIX_NORMAL.value} ), }, "optional": { "mask": ("MASK",), } } RETURN_TYPES = ("IMAGE",) FUNCTION = "blend" CATEGORY = "ImageBlender" def blend(self, base_image: torch.Tensor, blend_image: torch.Tensor, strength: float, blend_mode: str, mask: torch.Tensor = None) -> tuple: assert base_image.shape == blend_image.shape, "Base and blend images must have the same shape" assert base_image.shape[-1] == 3, "Input images must have 3 channels (RGB)" blend_function = self.blend_functions.get(BlendModes(blend_mode), lambda x, y: x) result = blend_function(base_image, blend_image) if mask is not None: # Ensure mask has the same number of channels as the images if mask.dim() == 3: mask = mask.unsqueeze(-1).expand(-1, -1, -1, base_image.shape[-1]) if mask.size() != base_image.size(): print(f"WARN: Mask size {mask.size()} is different from image size {base_image.size()}, mask is ignored") else: result = result * mask + base_image * (1 - mask) # Apply opacity result = result * strength + base_image * (1 - strength) # Normalize the result result = torch.clamp(result, 0, 1) return (result,) NODE_CLASS_MAPPINGS = { "ImageBlender": ImageBlender } NODE_DISPLAY_NAME_MAPPINGS = { "ImageBlender": "ImageBlender" }