Fioceen's picture
Fix node
8e6df85
raw
history blame
13 kB
import torch
from PIL import Image
import numpy as np
import os
import tempfile
from types import SimpleNamespace
from typing import Tuple
try:
from .image_postprocess import process_image
except Exception as e:
process_image = None
IMPORT_ERROR = str(e)
else:
IMPORT_ERROR = None
lut_extensions = ['png','npy','cube']
class NovaNodes:
"""
ComfyUI node: Full post-processing chain using process_image from image_postprocess
All augmentations with tunable parameters.
NOTE: Adjusted to match FOOLAI output:
- Returns an IMAGE as a single PyTorch tensor shaped (1, H, W, C), dtype=float32, values in [0.0, 1.0].
- Returns EXIF as a STRING (second output slot).
Added LUT support: two new node inputs:
- lut: STRING path to a LUT file (1D PNG 256x1, .npy, or .cube). Empty string -> disabled.
- lut_strength: FLOAT blend strength (0.0..1.0)
"""
@classmethod
def INPUT_TYPES(s):
# --- MODIFICATION: Rearranged inputs and updated defaults to match the reference image ---
return {
"required": {
"image": ("IMAGE",),
# Parameters (Manual Mode) - Order and defaults match the image
"noise_std_frac": ("FLOAT", {"default": 0.02, "min": 0.0, "max": 0.1, "step": 0.001}),
"clahe_clip": ("FLOAT", {"default": 2.00, "min": 0.5, "max": 10.0, "step": 0.1}),
"clahe_grid": ("INT", {"default": 8, "min": 2, "max": 32, "step": 1}),
"fourier_cutoff": ("FLOAT", {"default": 0.25, "min": 0.0, "max": 1.0, "step": 0.01}),
"apply_fourier_o": ("BOOLEAN", {"default": True}),
"fourier_strength": ("FLOAT", {"default": 0.90, "min": 0.0, "max": 1.0, "step": 0.01}),
"fourier_randomness": ("FLOAT", {"default": 0.05, "min": 0.0, "max": 0.5, "step": 0.01}),
"fourier_phase_perturb": ("FLOAT", {"default": 0.08, "min": 0.0, "max": 0.5, "step": 0.01}),
"fourier_radial_smooth": ("INT", {"default": 5, "min": 0, "max": 50, "step": 1}),
"fourier_mode": (["auto", "ref", "model"], {"default": "auto"}),
"fourier_alpha": ("FLOAT", {"default": 1.00, "min": 0.1, "max": 4.0, "step": 0.1}),
"perturb_mag_frac": ("FLOAT", {"default": 0.01, "min": 0.0, "max": 0.05, "step": 0.001}),
"enable_awb": ("BOOLEAN", {"default": True}),
"sim_camera": ("BOOLEAN", {"default": True}), # This corresponds to "Enable camera pipeline simulation"
"enable_lut": ("BOOLEAN", {"default": True}),
"lut": ("STRING", {"default": "X://insert/path/here(.png/.npy/.cube)", "vhs_path_extensions": lut_extensions}),
"lut_strength": ("FLOAT", {"default": 1.00, "min": 0.0, "max": 1.0, "step": 0.01}),
# Camera simulator options - Order and defaults match the image
"enable_bayer": ("BOOLEAN", {"default": True}),
"apply_jpeg_cycles_o": ("BOOLEAN", {"default": True}),
"jpeg_cycles": ("INT", {"default": 1, "min": 1, "max": 10, "step": 1}),
"jpeg_quality": ("INT", {"default": 88, "min": 10, "max": 100, "step": 1}),
"apply_vignette_o": ("BOOLEAN", {"default": True}),
"vignette_strength": ("FLOAT", {"default": 0.35, "min": 0.0, "max": 1.0, "step": 0.01}),
"apply_chromatic_aberration_o": ("BOOLEAN", {"default": True}),
"ca_shift": ("FLOAT", {"default": 1.20, "min": 0.0, "max": 5.0, "step": 0.1}),
"iso_scale": ("FLOAT", {"default": 1.00, "min": 0.1, "max": 16.0, "step": 0.1}),
"read_noise": ("FLOAT", {"default": 2.00, "min": 0.0, "max": 50.0, "step": 0.1}),
"hot_pixel_prob": ("FLOAT", {"default": 1e-7, "min": 0.0, "max": 1e-3, "step": 1e-7}),
"apply_banding_o": ("BOOLEAN", {"default": True}),
"banding_strength": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 1.0, "step": 0.01}),
"apply_motion_blur_o": ("BOOLEAN", {"default": True}),
"motion_blur_ksize": ("INT", {"default": 1, "min": 1, "max": 31, "step": 2}),
# Other options
"apply_exif_o": ("BOOLEAN", {"default": True}),
},
"optional": {
"awb_ref_image": ("IMAGE",),
"fft_ref_image": ("IMAGE",),
}
}
RETURN_TYPES = ("IMAGE", "STRING")
RETURN_NAMES = ("IMAGE", "EXIF")
FUNCTION = "process"
CATEGORY = "postprocessing"
def process(self, image,
noise_std_frac=0.02,
clahe_clip=2.0,
clahe_grid=8,
fourier_cutoff=0.25,
apply_fourier_o=True,
fourier_strength=0.9,
fourier_randomness=0.05,
fourier_phase_perturb=0.08,
fourier_radial_smooth=5,
fourier_mode="auto",
fourier_alpha=1.0,
perturb_mag_frac=0.01,
enable_awb=True,
sim_camera=True,
enable_lut=True,
lut="",
lut_strength=1.0,
enable_bayer=True,
apply_jpeg_cycles_o=True,
jpeg_cycles=1,
jpeg_quality=88,
apply_vignette_o=True,
vignette_strength=0.35,
apply_chromatic_aberration_o=True,
ca_shift=1.20,
iso_scale=1.0,
read_noise=2.0,
hot_pixel_prob=1e-7,
apply_banding_o=True,
banding_strength=0.0,
apply_motion_blur_o=True,
motion_blur_ksize=1,
apply_exif_o=True,
awb_ref_image=None,
fft_ref_image=None
):
if process_image is None:
raise ImportError(f"Could not import process_image function: {IMPORT_ERROR}")
tmp_files = []
def to_pil_from_any(inp):
"""Convert a torch tensor / numpy array of many shapes into a PIL RGB Image."""
if isinstance(inp, torch.Tensor):
arr = inp.detach().cpu().numpy()
else:
arr = np.asarray(inp)
if arr.ndim == 4 and arr.shape[0] == 1:
arr = arr[0]
if arr.ndim == 3 and arr.shape[0] in (1, 3):
arr = np.transpose(arr, (1, 2, 0))
if arr.ndim == 2:
arr = arr[:, :, None]
if arr.ndim != 3:
raise TypeError(f"Cannot convert array to HWC image, final ndim={arr.ndim}, shape={arr.shape}")
if np.issubdtype(arr.dtype, np.floating):
if arr.max() <= 1.0:
arr = (arr * 255.0).clip(0, 255).astype(np.uint8)
else:
arr = np.clip(arr, 0, 255).astype(np.uint8)
else:
arr = arr.astype(np.uint8)
if arr.shape[2] == 1:
arr = np.repeat(arr, 3, axis=2)
return Image.fromarray(arr)
try:
# ---- Input image -> temporary input file ----
pil_img = to_pil_from_any(image[0])
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_input:
input_path = tmp_input.name
pil_img.save(input_path)
tmp_files.append(input_path)
# ---- AWB reference image if present ----
awb_ref_path = None
if awb_ref_image is not None:
pil_ref_awb = to_pil_from_any(awb_ref_image[0])
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_ref_awb:
awb_ref_path = tmp_ref_awb.name
pil_ref_awb.save(awb_ref_path)
tmp_files.append(awb_ref_path)
# ---- FFT reference image if present ----
fft_ref_path = None
if fft_ref_image is not None:
pil_ref_fft = to_pil_from_any(fft_ref_image[0])
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_ref_fft:
fft_ref_path = tmp_ref_fft.name
pil_ref_fft.save(fft_ref_path)
tmp_files.append(fft_ref_path)
# ---- Output path ----
with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp_output:
output_path = tmp_output.name
tmp_files.append(output_path)
# Prepare args for process_image
args = SimpleNamespace(
input=input_path,
output=output_path,
awb=enable_awb, # Explicit AWB flag
ref=awb_ref_path,
noise_std=noise_std_frac,
hot_pixel_prob=hot_pixel_prob,
perturb=perturb_mag_frac,
clahe_clip=clahe_clip,
tile=clahe_grid,
fstrength=fourier_strength if apply_fourier_o else 0.0,
randomness=fourier_randomness,
phase_perturb=fourier_phase_perturb,
fft_alpha=fourier_alpha,
radial_smooth=fourier_radial_smooth,
fft_mode=fourier_mode,
fft_ref=fft_ref_path,
vignette_strength=vignette_strength if apply_vignette_o else 0.0,
chroma_strength=ca_shift if apply_chromatic_aberration_o else 0.0,
banding_strength=banding_strength if apply_banding_o else 0.0,
motion_blur_kernel=motion_blur_ksize if apply_motion_blur_o else 1,
jpeg_cycles=jpeg_cycles if apply_jpeg_cycles_o else 1,
jpeg_qmin=jpeg_quality,
jpeg_qmax=96, # As per image range
sim_camera=sim_camera,
no_no_bayer=not enable_bayer, # FIX: Inverted logic corrected
iso_scale=iso_scale,
read_noise=read_noise,
seed=None, # Seed is not user-configurable in this version
cutoff=fourier_cutoff,
lut=(lut if enable_lut and lut != "" else None),
lut_strength=lut_strength,
)
# ---- Run the processing function ----
process_image(input_path, output_path, args)
# ---- Load result (force RGB) ----
output_img = Image.open(output_path).convert("RGB")
img_out = np.array(output_img)
# ---- EXIF insertion (optional) ----
new_exif = ""
if apply_exif_o:
try:
output_img_with_exif, new_exif = self._add_fake_exif(output_img)
output_img = output_img_with_exif
img_out = np.array(output_img.convert("RGB"))
except Exception:
new_exif = ""
# ---- Convert to FOOLAI-style tensor: (1, H, W, C), float32 in [0,1] ----
img_float = img_out.astype(np.float32) / 255.0
tensor_out = torch.from_numpy(img_float).to(dtype=torch.float32).unsqueeze(0)
tensor_out = torch.clamp(tensor_out, 0.0, 1.0)
return (tensor_out, new_exif)
finally:
for p in tmp_files:
try:
os.unlink(p)
except Exception:
pass
def _add_fake_exif(self, img: Image.Image) -> Tuple[Image.Image, str]:
"""Insert random but realistic camera EXIF metadata."""
import random
import io
try:
import piexif
except Exception:
raise
exif_dict = {
"0th": {
piexif.ImageIFD.Make: random.choice(["Canon", "Nikon", "Sony", "Fujifilm", "Olympus", "Leica"]),
piexif.ImageIFD.Model: random.choice([
"EOS 5D Mark III", "D850", "Alpha 7R IV", "X-T4", "OM-D E-M1 Mark III", "Q2"
]),
piexif.ImageIFD.Software: "Adobe Lightroom",
},
"Exif": {
piexif.ExifIFD.FNumber: (random.randint(10, 22), 10),
piexif.ExifIFD.ExposureTime: (1, random.randint(60, 4000)),
piexif.ExifIFD.ISOSpeedRatings: random.choice([100, 200, 400, 800, 1600, 3200]),
piexif.ExifIFD.FocalLength: (random.randint(24, 200), 1),
},
}
exif_bytes = piexif.dump(exif_dict)
output = io.BytesIO()
img.save(output, format="JPEG", exif=exif_bytes)
output.seek(0)
return (Image.open(output), str(exif_bytes))
# -------------
# Registration
# -------------
NODE_CLASS_MAPPINGS = {
"NovaNodes": NovaNodes,
}
NODE_DISPLAY_NAME_MAPPINGS = {
"NovaNodes": "Image Postprocess (NOVA NODES)",
}