xiaden
7d82aa22e6
Sub-packages: - edge/ — Canny, Lineart, LineartAnime (3 nodes) - restore/ — SCUNet, NAFNet, Adversarial (3 nodes) - compare/ — CCIP, LPIPS (2 nodes) - validate/ — Classify (11 ops), Check (5 ops), Metric (2 ops) - tagging/ — WD14, DeepDanbooru, DeepGelbooru, MLDanbooru, Camie, PixAI (6 nodes) Root nodes: - ImgUtilsOCR — PaddleOCR with heat/box threshold - ImgUtilsDetect — 10 detection types + confidence knob - ImgUtilsPose — DWPose + auto_detect toggle - ImgUtilsSegment — isnetis + scale knob - ImgUtilsTags — tag string processing - ImgUtilsCDC — upscaling + tile_size - ImgUtilsCensor — NSFW censoring with method/confidence/body-part toggles - ImgUtilsAlign — resize to max dimension - ImgUtilsSqueeze — transparency-based auto-crop Every node exposes its native function parameters as knobs. Removed mega-nodes: transform (11 ops), compare (3 ops), validate (18 ops).
70 lines
2.4 KiB
Python
70 lines
2.4 KiB
Python
"""
|
|
Node 6: ImgUtilsSegment — "Segment character"
|
|
|
|
Segments character from background using isnetis (anime character segmentation).
|
|
Outputs both a mask image and the segmented RGBA image with transparent background.
|
|
|
|
Uses:
|
|
imgutils.segment.isnetis.segment_rgba_with_isnetis() -> (mask: PIL, rgba_image: PIL)
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from comfy_api.latest import io, ui
|
|
|
|
from .utils import comfy_to_pil, pil_to_comfy
|
|
|
|
|
|
class ImgUtilsSegment(io.ComfyNode):
|
|
"""Segment character from background using isnetis."""
|
|
|
|
@classmethod
|
|
def define_schema(cls) -> io.Schema:
|
|
return io.Schema(
|
|
node_id="ImgUtilsSegment",
|
|
display_name="Imgutils Segment",
|
|
category="imgutils/segment",
|
|
description=(
|
|
"Segment anime character from background using isnetis. "
|
|
"Outputs the alpha mask and an RGBA image with transparent background."
|
|
),
|
|
search_aliases=[
|
|
"segment", "isnetis", "mask", "character",
|
|
"background", "remove bg", "cutout", "alpha",
|
|
],
|
|
inputs=[
|
|
io.Image.Input("image", tooltip="Input image containing character to segment"),
|
|
io.Int.Input(
|
|
"scale", default=1024, min=256, max=2048, step=128,
|
|
tooltip="Processing resolution. Higher = more accurate, slower. Default 1024.",
|
|
),
|
|
],
|
|
outputs=[
|
|
io.Mask.Output(display_name="mask"),
|
|
io.Image.Output(display_name="segmented_rgba"),
|
|
],
|
|
)
|
|
|
|
@classmethod
|
|
def execute(cls, image, scale=1024) -> io.NodeOutput:
|
|
"""
|
|
Run character segmentation and return mask + RGBA image.
|
|
|
|
Args:
|
|
image: ComfyUI IMAGE tensor (B,H,W,C), float32, [0,1]
|
|
|
|
Returns:
|
|
NodeOutput with (mask_tensor, segmented_rgba_tensor)
|
|
"""
|
|
from imgutils.segment import segment_rgba_with_isnetis
|
|
|
|
pil_image = comfy_to_pil(image.numpy() if hasattr(image, "numpy") else image)
|
|
|
|
mask_pil, rgba_pil = segment_rgba_with_isnetis(pil_image, scale=int(scale))
|
|
|
|
# pil_to_comfy handles grayscale "L" mode (mask) and "RGBA" mode
|
|
mask_tensor = pil_to_comfy(mask_pil)
|
|
rgba_tensor = pil_to_comfy(rgba_pil)
|
|
|
|
return io.NodeOutput(mask_tensor, rgba_tensor, ui=ui.PreviewImage(rgba_tensor, cls=cls))
|