Superside ComfyUI Nodes
6ebd0aa6cd
Ported and curated from the internal fal-flux-nodes package: dropped 3 redundant nodes (Nano Banana v1, AnyVLM, SeedVR v1) that were fully superseded by newer variants already in the set. Every node now takes an explicit api_key STRING input instead of reading from a config file or environment variable, so no key is ever baked into this repo.
138 lines
5.1 KiB
Python
138 lines
5.1 KiB
Python
import io
|
|
import logging
|
|
|
|
import numpy as np
|
|
import torch
|
|
from PIL import Image, ImageFilter
|
|
|
|
from .base_node import (
|
|
SupersideFalNode,
|
|
ImageProcessingMixin,
|
|
APIClientMixin,
|
|
API_KEY_INPUT_SPEC,
|
|
)
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class BriaBackgroundStandardizerNode(
|
|
SupersideFalNode, ImageProcessingMixin, APIClientMixin
|
|
):
|
|
"""
|
|
Bria Background Standardizer Node: Cut out the subject with
|
|
fal-ai/bria/background/remove, then composite it locally onto a solid
|
|
hex-color canvas.
|
|
|
|
Unlike prompt-driven background replacement, this node never regenerates
|
|
the subject. The cutout is pixel-identical to the source, and the
|
|
background is an exact hex color instead of a diffusion model's
|
|
interpretation of a color description - this makes it suitable for
|
|
batch-homogenizing avatar backgrounds without any quality drift.
|
|
"""
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"image": ("IMAGE",),
|
|
"hex_color": (
|
|
"STRING",
|
|
{
|
|
"default": "#FFFFFF",
|
|
"placeholder": "#RRGGBB (e.g. #F5F5F5)",
|
|
},
|
|
),
|
|
"api_key": API_KEY_INPUT_SPEC,
|
|
},
|
|
"optional": {
|
|
"edge_feather": (
|
|
"INT",
|
|
{
|
|
"default": 0,
|
|
"min": 0,
|
|
"max": 15,
|
|
"tooltip": "Softens the cutout edge by this many pixels to avoid a jagged silhouette. 0 = pixel-perfect edge, no softening.",
|
|
},
|
|
),
|
|
"sync_mode": ("BOOLEAN", {"default": False}),
|
|
},
|
|
"hidden": {
|
|
"unique_id": "UNIQUE_ID",
|
|
"extra_pnginfo": "EXTRA_PNGINFO",
|
|
},
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE", "STRING")
|
|
RETURN_NAMES = ("image", "info")
|
|
FUNCTION = "generate"
|
|
OUTPUT_NODE = True
|
|
DESCRIPTION = (
|
|
"Standardize an avatar's background to an exact hex color. Removes the "
|
|
"background with Bria RMBG 2.0 (fal-ai/bria/background/remove) and "
|
|
"composites the untouched subject onto a solid color locally - the "
|
|
"subject is never regenerated, only the background pixels change."
|
|
)
|
|
|
|
@staticmethod
|
|
def _parse_hex_color(hex_color):
|
|
value = hex_color.strip().lstrip("#")
|
|
if len(value) == 3:
|
|
value = "".join(ch * 2 for ch in value)
|
|
if len(value) != 6 or any(ch not in "0123456789abcdefABCDEF" for ch in value):
|
|
raise ValueError(
|
|
f"Invalid hex_color '{hex_color}'. Use a format like '#FFFFFF' or '#FFF'."
|
|
)
|
|
return tuple(int(value[i : i + 2], 16) for i in (0, 2, 4))
|
|
|
|
def generate(self, image, hex_color, api_key, unique_id=None, extra_pnginfo=None, **kwargs):
|
|
try:
|
|
client = self.get_client(api_key)
|
|
rgb = self._parse_hex_color(hex_color)
|
|
|
|
image_url = self.upload_image(client, image)
|
|
arguments = {
|
|
"image_url": image_url,
|
|
"sync_mode": kwargs.get("sync_mode", False),
|
|
}
|
|
result = self.call_api(client, "fal-ai/bria/background/remove", arguments)
|
|
|
|
cutout_data = result.get("image")
|
|
if not isinstance(cutout_data, dict) or "url" not in cutout_data:
|
|
raise RuntimeError("No cutout image was returned by the background removal API.")
|
|
|
|
cutout_url = cutout_data["url"]
|
|
cutout_bytes = self._load_image_bytes_from_url(cutout_url)
|
|
cutout = Image.open(io.BytesIO(cutout_bytes)).convert("RGBA")
|
|
|
|
edge_feather = kwargs.get("edge_feather", 0)
|
|
if edge_feather:
|
|
alpha = cutout.split()[-1].filter(ImageFilter.GaussianBlur(edge_feather))
|
|
cutout.putalpha(alpha)
|
|
|
|
background = Image.new("RGBA", cutout.size, rgb + (255,))
|
|
composed = Image.alpha_composite(background, cutout).convert("RGB")
|
|
|
|
img_np = np.array(composed).astype(np.float32) / 255.0
|
|
img_tensor = torch.from_numpy(img_np).unsqueeze(0)
|
|
|
|
info = f"background=#{''.join(f'{c:02X}' for c in rgb)}; source_cutout={cutout_url}"
|
|
|
|
if unique_id is not None and extra_pnginfo is not None:
|
|
if (
|
|
isinstance(extra_pnginfo, list)
|
|
and isinstance(extra_pnginfo[0], dict)
|
|
and "workflow" in extra_pnginfo[0]
|
|
):
|
|
workflow = extra_pnginfo[0]["workflow"]
|
|
node = next(
|
|
(x for x in workflow["nodes"] if str(x["id"]) == str(unique_id)),
|
|
None,
|
|
)
|
|
if node:
|
|
node["widgets_values"] = [info]
|
|
|
|
return {"ui": {"text": [info]}, "result": (img_tensor, info)}
|
|
except Exception as e:
|
|
logger.error(f"Bria background standardizer failed: {str(e)}")
|
|
raise RuntimeError(f"Bria background standardizer failed: {str(e)}") from e
|