HakoniwaandClaude Opus 5.5
90e1371b76
- textbbox was measured at the patch position and that position was then subtracted again, so every number was drawn inside patch 0. Measure at the origin and subtract only the glyph offset. - Fall back to DejaVu Sans / Helvetica, then Pillow's scalable default font, so font_size works where Arial is not installed. - Add regression tests for label placement and font size. - Add the Linux/Windows CPU test workflow that the README already describes. - Add .gitignore for __pycache__, bump version to 1.2.1, update README and VALIDATION.md. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K5PqyHqy9udCb9y3WWmeTx
433 lines
20 KiB
Python
433 lines
20 KiB
Python
import json
|
|
import random
|
|
import numpy as np
|
|
import torch
|
|
import torch.nn.functional as F
|
|
from PIL import Image, ImageDraw, ImageFont
|
|
from sklearn.cluster import KMeans
|
|
|
|
|
|
def _validate_images(images):
|
|
if (not isinstance(images, torch.Tensor) or images.ndim != 4
|
|
or images.shape[-1] != 3 or any(size == 0 for size in images.shape)):
|
|
raise ValueError("images must have shape [batch, height, width, 3] (RGB).")
|
|
if not images.is_floating_point() or not torch.isfinite(images).all():
|
|
raise ValueError("images must contain finite floating-point RGB values.")
|
|
|
|
|
|
def _validate_palette(palette):
|
|
if (not isinstance(palette, torch.Tensor) or palette.ndim != 2
|
|
or palette.shape[0] == 0 or palette.shape[1] != 3):
|
|
raise ValueError("palette must have shape [color_count, 3] and contain at least one color.")
|
|
if (not torch.isfinite(palette).all()
|
|
or ((palette < 0) | (palette > 1)).any()):
|
|
raise ValueError("palette RGB values must be finite and between 0 and 1.")
|
|
|
|
|
|
def _prepare_indexed_inputs(images, palette, index_maps):
|
|
_validate_images(images)
|
|
_validate_palette(palette)
|
|
if not isinstance(index_maps, torch.Tensor) or index_maps.shape != images.shape[:3]:
|
|
raise ValueError("index_maps must match images: [batch, height, width].")
|
|
if (not torch.isfinite(index_maps).all() or (index_maps < 0).any()
|
|
or (index_maps >= len(palette)).any()
|
|
or (index_maps.is_floating_point() and (index_maps != index_maps.round()).any())):
|
|
raise ValueError("index_maps must contain integer palette indices; connect ColorshiftColor's index_maps output.")
|
|
return (palette.to(device=images.device, dtype=torch.float32),
|
|
index_maps.to(device=images.device, dtype=torch.long))
|
|
|
|
|
|
def _parse_operations(operations):
|
|
if not isinstance(operations, str):
|
|
raise ValueError("operations must be a JSON list.")
|
|
try:
|
|
# Shipped workflows disable example operations with a leading ## line comment.
|
|
active_lines = "\n".join(line for line in operations.splitlines()
|
|
if not line.lstrip().startswith("##"))
|
|
result = json.loads(active_lines.strip() or "[]")
|
|
except ValueError as exc:
|
|
raise ValueError(f"operations contains invalid JSON: {exc}") from exc
|
|
if not isinstance(result, list) or any(not isinstance(op, dict) for op in result):
|
|
raise ValueError("operations must be a JSON list of objects.")
|
|
return result
|
|
|
|
|
|
def _color_index(value, count, name):
|
|
if type(value) is not int or not 0 <= value < count:
|
|
raise ValueError(f"{name} must be an integer between 0 and {count - 1}.")
|
|
return value
|
|
|
|
|
|
def _color_triplet(value, palette, name):
|
|
if (not isinstance(value, list) or len(value) != 3
|
|
or any(type(component) not in (int, float) for component in value)):
|
|
raise ValueError(f"{name} must contain three numbers between 0 and 1.")
|
|
color = palette.new_tensor(value)
|
|
if not torch.isfinite(color).all() or ((color < 0) | (color > 1)).any():
|
|
raise ValueError(f"{name} must contain three finite numbers between 0 and 1.")
|
|
return color
|
|
|
|
|
|
def _load_font(font_size):
|
|
# Arial exists on Windows only; fall back to other common fonts, then Pillow's
|
|
# scalable default (Pillow >= 10.1) so font_size is honored on macOS/Linux too.
|
|
for name in ("arial.ttf", "Arial.ttf", "DejaVuSans.ttf", "Helvetica.ttc"):
|
|
try:
|
|
return ImageFont.truetype(name, font_size)
|
|
except OSError:
|
|
pass
|
|
try:
|
|
return ImageFont.load_default(size=font_size)
|
|
except TypeError:
|
|
return ImageFont.load_default()
|
|
|
|
|
|
class ColorshiftColorNode:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"images": ("IMAGE",),
|
|
"color_count": ("INT", {"default": 8, "min": 2, "max": 64}),
|
|
},
|
|
"optional": {
|
|
"lock_masks": ("MASK", {"default": None}),
|
|
"palette_override": ("PALETTE", {"default": None}),
|
|
"font_size": ("INT", {"default": 20, "min": 10, "max": 50}),
|
|
"sampling_rate": ("FLOAT", {"default": 0.25, "min": 0.01, "max": 0.5}),
|
|
"n_init": ("INT", {"default": 3, "min": 1, "max": 10}),
|
|
"seed": ("INT", {"default": 0, "min": 0, "max": 0xFFFFFFFFFFFFFFFF}),
|
|
},
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE", "PALETTE", "MASK", "IMAGE")
|
|
RETURN_NAMES = ("images", "palette", "index_maps", "palette_preview")
|
|
FUNCTION = "process"
|
|
CATEGORY = "Image/Color"
|
|
|
|
@torch.no_grad()
|
|
def process(self, images, color_count, lock_masks=None, palette_override=None, font_size=20, sampling_rate=0.25, n_init=3, seed=0):
|
|
_validate_images(images)
|
|
if type(color_count) is not int or not 2 <= color_count <= 64:
|
|
raise ValueError("color_count must be an integer between 2 and 64.")
|
|
if not 0.01 <= sampling_rate <= 0.5:
|
|
raise ValueError("sampling_rate must be between 0.01 and 0.5.")
|
|
if type(n_init) is not int or not 1 <= n_init <= 10:
|
|
raise ValueError("n_init must be an integer between 1 and 10.")
|
|
batch_size, height, width, channels = images.shape
|
|
pixels = images.reshape(batch_size, -1, 3)
|
|
|
|
if palette_override is not None:
|
|
_validate_palette(palette_override)
|
|
palette = palette_override.to(device=images.device, dtype=torch.float32).clone()
|
|
labels = self._match_palette_batch(pixels, palette)
|
|
else:
|
|
# Sample original pixels so resizing cannot invent intermediate colors.
|
|
rng = np.random.default_rng(seed)
|
|
flat_pixels = pixels.reshape(-1, 3)
|
|
sample_count = min(len(flat_pixels), 262144, max(color_count, int(len(flat_pixels) * sampling_rate)))
|
|
random_indices = torch.as_tensor(
|
|
rng.choice(len(flat_pixels), size=sample_count, replace=False), device=images.device)
|
|
samples = flat_pixels[random_indices].float().clamp(0, 1).cpu().numpy()
|
|
unique_colors, weights = np.unique(samples, axis=0, return_counts=True)
|
|
cluster_count = min(color_count, len(unique_colors))
|
|
if cluster_count == len(unique_colors):
|
|
centers = unique_colors
|
|
else:
|
|
kmeans = KMeans(n_clusters=cluster_count, random_state=int(seed) % (2**32), n_init=n_init)
|
|
kmeans.fit(unique_colors, sample_weight=weights)
|
|
centers = kmeans.cluster_centers_
|
|
palette = torch.as_tensor(centers, device=images.device, dtype=torch.float32).clamp(0, 1)
|
|
labels = self._match_palette_batch(pixels, palette)
|
|
|
|
# ----- パレットの並び順を各クラスタの占有率が高い順に変更 -----
|
|
# 全ラベルを1次元にまとめ、各クラスタの出現回数を計算
|
|
all_labels = labels.flatten()
|
|
counts = torch.bincount(all_labels, minlength=len(palette))
|
|
# 出現回数の降順に並べ替えたときの各クラスタの元のインデックス
|
|
sorted_order = torch.argsort(counts, descending=True, stable=True)
|
|
# mapping: 旧インデックス -> 新インデックス を作成
|
|
mapping = torch.empty_like(sorted_order)
|
|
mapping[sorted_order] = torch.arange(sorted_order.size(0), device=sorted_order.device)
|
|
# パレットを並び替え、各画素のラベルを新しい順序に置換
|
|
palette = palette[sorted_order]
|
|
labels = mapping[labels]
|
|
# ---------------------------------------------------------------
|
|
|
|
preview_img = self.generate_palette_preview(palette, font_size)
|
|
index_maps = labels.reshape(batch_size, height, width)
|
|
processed_images_tensor = palette[labels].reshape(images.shape).to(images.dtype)
|
|
if lock_masks is not None:
|
|
if not isinstance(lock_masks, torch.Tensor) or lock_masks.ndim not in (2, 3):
|
|
raise ValueError("lock_masks must have shape [height, width] or [batch, height, width].")
|
|
if any(size == 0 for size in lock_masks.shape) or not torch.isfinite(lock_masks).all():
|
|
raise ValueError("lock_masks must be nonempty and contain finite values.")
|
|
masks = lock_masks.to(device=images.device, dtype=torch.float32)
|
|
if masks.ndim == 2:
|
|
masks = masks.unsqueeze(0)
|
|
if masks.shape[1:] != (height, width):
|
|
masks = F.interpolate(masks.unsqueeze(1), size=(height, width), mode="bilinear", align_corners=False).squeeze(1)
|
|
batch_indices = torch.arange(batch_size, device=images.device).clamp(max=len(masks) - 1)
|
|
masks = masks[batch_indices].clamp(0, 1).unsqueeze(-1).to(images.dtype)
|
|
processed_images_tensor = images * masks + processed_images_tensor * (1 - masks)
|
|
return (processed_images_tensor, palette, index_maps, preview_img)
|
|
|
|
def generate_palette_preview(self, palette, font_size):
|
|
if isinstance(palette, torch.Tensor):
|
|
colors = palette.detach().float().cpu().numpy()
|
|
else:
|
|
colors = palette
|
|
|
|
patch_size = 100
|
|
cols = 8
|
|
rows = (len(colors) + cols - 1) // cols
|
|
|
|
img = Image.new("RGB", (cols * patch_size, rows * patch_size), (40, 40, 40))
|
|
draw = ImageDraw.Draw(img)
|
|
|
|
font = _load_font(font_size)
|
|
|
|
for i, color in enumerate(colors):
|
|
x = (i % cols) * patch_size
|
|
y = (i // cols) * patch_size
|
|
|
|
draw.rectangle([x, y, x + patch_size - 1, y + patch_size - 1], fill=tuple((color * 255).clip(0, 255).astype(int)))
|
|
text = str(i)
|
|
# Measure at the origin so the bbox offset is relative to the glyph, not the patch.
|
|
left, top, right, bottom = draw.textbbox((0, 0), text, font=font)
|
|
text_x = x + (patch_size - (right - left)) // 2 - left
|
|
text_y = y + (patch_size - (bottom - top)) // 2 - top
|
|
text_color = "black" if np.dot(color, [0.2126, 0.7152, 0.0722]) > 0.5 else "white"
|
|
draw.text((text_x, text_y), text, fill=text_color, font=font)
|
|
|
|
img_tensor = torch.from_numpy(np.array(img).astype(np.float32) / 255.0).unsqueeze(0)
|
|
return img_tensor
|
|
|
|
def _match_palette_batch(self, pixels, palette):
|
|
# Bound the temporary distance matrix, including for large override palettes.
|
|
flat_pixels = pixels.reshape(-1, 3)
|
|
palette = palette.to(device=pixels.device, dtype=torch.float32)
|
|
chunk_size = max(1, min(65536, 4_194_304 // len(palette)))
|
|
labels = torch.empty(len(flat_pixels), device=pixels.device, dtype=torch.long)
|
|
for start in range(0, len(flat_pixels), chunk_size):
|
|
chunk = flat_pixels[start:start + chunk_size].float()
|
|
distances = torch.cdist(chunk, palette, compute_mode="donot_use_mm_for_euclid_dist")
|
|
labels[start:start + chunk_size] = distances.argmin(dim=-1)
|
|
return labels.reshape(pixels.shape[:-1])
|
|
|
|
class PaletteEditorNode:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"images": ("IMAGE",),
|
|
"palette": ("PALETTE",),
|
|
"index_maps": ("MASK",),
|
|
},
|
|
"optional": {
|
|
"seed": ("INT", {"default": 0, "min": 0, "max": 0xFFFFFFFFFFFFFFFF}),
|
|
"hue_random_enable": ("BOOLEAN", {"default": False}),
|
|
"hue_shift": ("FLOAT", {"default": 0.0, "min": -180.0, "max": 180.0}),
|
|
"saturation_random_enable": ("BOOLEAN", {"default": False}),
|
|
"saturation_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 5.0}),
|
|
"value_random_enable": ("BOOLEAN", {"default": False}),
|
|
"value_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 5.0}),
|
|
"lock_color_num": ("STRING", {"default": "0,", "placeholder": "例: 0,1,2,3,4"}),
|
|
"mask_enable": ("BOOLEAN", {"default": False}),
|
|
"invert_mask": ("BOOLEAN", {"default": False}),
|
|
"operations": ("STRING", {"default": "[]", "multiline": True}),
|
|
},
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE", "PALETTE", "MASK")
|
|
RETURN_NAMES = ("images", "palette", "mask")
|
|
FUNCTION = "process_palette"
|
|
CATEGORY = "Image/Color"
|
|
|
|
def rgb_to_hsv(self, rgb):
|
|
r, g, b = rgb.unbind(-1)
|
|
max_rgb, argmax_rgb = rgb.max(-1)
|
|
min_rgb, _ = rgb.min(-1)
|
|
delta = max_rgb - min_rgb
|
|
|
|
h = torch.zeros_like(r)
|
|
s = torch.zeros_like(r)
|
|
v = max_rgb
|
|
|
|
non_zero_delta = delta != 0
|
|
h[non_zero_delta] = torch.where(
|
|
argmax_rgb[non_zero_delta] == 0,
|
|
(g[non_zero_delta] - b[non_zero_delta]) / delta[non_zero_delta],
|
|
torch.where(
|
|
argmax_rgb[non_zero_delta] == 1,
|
|
2 + (b[non_zero_delta] - r[non_zero_delta]) / delta[non_zero_delta],
|
|
4 + (r[non_zero_delta] - g[non_zero_delta]) / delta[non_zero_delta],
|
|
),
|
|
)
|
|
h = (h / 6.0) % 1.0
|
|
s[non_zero_delta] = delta[non_zero_delta] / max_rgb[non_zero_delta]
|
|
|
|
return torch.stack((h, s, v), dim=-1)
|
|
|
|
def hsv_to_rgb(self, hsv):
|
|
h, s, v = hsv.unbind(-1)
|
|
|
|
c = v * s
|
|
x = c * (1 - torch.abs((h * 6) % 2 - 1))
|
|
m = v - c
|
|
|
|
rgb_prime = torch.zeros_like(hsv)
|
|
|
|
h_category = (h * 6).long() % 6
|
|
|
|
rgb_prime[..., 0] = torch.where(
|
|
(h_category == 0) | (h_category == 5),
|
|
c,
|
|
torch.where((h_category == 1) | (h_category == 4), x, 0),
|
|
)
|
|
rgb_prime[..., 1] = torch.where(
|
|
(h_category == 1) | (h_category == 2),
|
|
c,
|
|
torch.where((h_category == 0) | (h_category == 3), x, 0),
|
|
)
|
|
rgb_prime[..., 2] = torch.where(
|
|
(h_category == 3) | (h_category == 4),
|
|
c,
|
|
torch.where((h_category == 2) | (h_category == 5), x, 0),
|
|
)
|
|
|
|
rgb = rgb_prime + m[..., None]
|
|
return rgb
|
|
|
|
@torch.no_grad()
|
|
def process_palette(self, images, palette, index_maps, seed=0, hue_random_enable=False, hue_shift=0.0, saturation_random_enable=False, saturation_scale=1.0, value_random_enable=False, value_scale=1.0, lock_color_num="0,", mask_enable=False, invert_mask=False, operations="[]"):
|
|
palette, index_maps = _prepare_indexed_inputs(images, palette, index_maps)
|
|
color_mask = torch.zeros(len(palette), dtype=torch.bool, device=images.device)
|
|
if mask_enable:
|
|
try:
|
|
indices = [int(idx.strip()) for idx in lock_color_num.split(",") if idx.strip()]
|
|
except ValueError as exc:
|
|
raise ValueError("lock_color_num must contain comma-separated integer color indices.") from exc
|
|
for idx in indices:
|
|
color_mask[_color_index(idx, len(palette), "lock_color_num")] = True
|
|
if invert_mask:
|
|
color_mask = ~color_mask
|
|
|
|
rng = random.Random(seed)
|
|
current_hue = rng.uniform(-180, 180) if hue_random_enable else hue_shift
|
|
current_saturation = rng.uniform(0, 5) if saturation_random_enable else saturation_scale
|
|
current_value = rng.uniform(0, 5) if value_random_enable else value_scale
|
|
|
|
modified_palette = self.edit_palette(palette, operations, color_mask, current_hue, current_saturation, current_value)
|
|
|
|
pixel_mask = color_mask[index_maps]
|
|
new_pixels = modified_palette[index_maps].to(images.dtype)
|
|
processed_images = torch.where(pixel_mask.unsqueeze(-1), images, new_pixels)
|
|
return (processed_images, modified_palette, pixel_mask.float())
|
|
|
|
def edit_palette(self, palette, operations, mask, hue_shift, saturation_scale, value_scale):
|
|
_validate_palette(palette)
|
|
palette = palette.clone()
|
|
mask = mask.to(device=palette.device, dtype=torch.bool)
|
|
hsv_palette = self.rgb_to_hsv(palette)
|
|
|
|
hsv_palette[~mask, 0] = (hsv_palette[~mask, 0] + hue_shift / 360.0) % 1.0
|
|
hsv_palette[~mask, 1] = torch.clamp(hsv_palette[~mask, 1] * saturation_scale, 0, 1)
|
|
hsv_palette[~mask, 2] = torch.clamp(hsv_palette[~mask, 2] * value_scale, 0, 1)
|
|
|
|
modified_palette = self.hsv_to_rgb(hsv_palette).clamp(0, 1)
|
|
modified_palette[mask] = palette[mask]
|
|
for op in _parse_operations(operations):
|
|
idx = _color_index(op.get("index"), len(palette), "operations.index")
|
|
if "color" not in op and "hsv" not in op:
|
|
raise ValueError("Each palette operation needs color or hsv.")
|
|
rgb = _color_triplet(op["color"], palette, "operations.color") if "color" in op else None
|
|
hsv = _color_triplet(op["hsv"], palette, "operations.hsv") if "hsv" in op else None
|
|
if rgb is not None:
|
|
modified_palette[idx] = rgb
|
|
if hsv is not None:
|
|
modified_palette[idx] = self.hsv_to_rgb(hsv.reshape(1, 3))[0]
|
|
|
|
return modified_palette
|
|
|
|
class CsCFill:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"images": ("IMAGE",),
|
|
"palette": ("PALETTE",),
|
|
"index_maps": ("MASK",),
|
|
},
|
|
"optional": {
|
|
"operations": ("STRING", {"default": "[]", "multiline": True}),
|
|
},
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE", "IMAGE")
|
|
RETURN_NAMES = ("filled_image", "mask_image")
|
|
FUNCTION = "process"
|
|
CATEGORY = "Image/Color"
|
|
|
|
@torch.no_grad()
|
|
def process(self, images, palette, index_maps, operations="[]"):
|
|
palette, index_maps = _prepare_indexed_inputs(images, palette, index_maps)
|
|
batch_size, height, width, channels = images.shape
|
|
ops = _parse_operations(operations)
|
|
pairs = [(_color_index(op.get("A"), len(palette), "operations.A"),
|
|
_color_index(op.get("B"), len(palette), "operations.B")) for op in ops]
|
|
if not ops:
|
|
pairs = self._compute_auto_pairs(palette)
|
|
|
|
# Resolve the palette mapping first; duplicate B entries use the last operation.
|
|
replacements = torch.arange(len(palette), device=images.device)
|
|
selected_colors = torch.zeros(len(palette), device=images.device, dtype=torch.bool)
|
|
for a_idx, b_idx in pairs:
|
|
replacements[b_idx] = a_idx
|
|
selected_colors[b_idx] = a_idx != b_idx
|
|
selected = selected_colors[index_maps]
|
|
filled_image = torch.where(selected.unsqueeze(-1),
|
|
palette[replacements[index_maps]].to(images.dtype), images)
|
|
mask_image = images.new_zeros((batch_size, height, width, 4))
|
|
mask_image[..., :3] = torch.where(selected.unsqueeze(-1), palette[index_maps].to(images.dtype), 0)
|
|
mask_image[..., 3] = selected.to(images.dtype)
|
|
return (filled_image, mask_image)
|
|
|
|
def _compute_auto_pairs(self, palette):
|
|
palette_np = palette.detach().cpu().numpy()
|
|
n = palette_np.shape[0]
|
|
pair_list = []
|
|
for i in range(n):
|
|
for j in range(i + 1, n):
|
|
dist = np.linalg.norm(palette_np[i] - palette_np[j])
|
|
pair_list.append((i, j, dist))
|
|
pair_list.sort(key=lambda x: x[2])
|
|
|
|
paired = set()
|
|
auto_pairs = []
|
|
for i, j, dist in pair_list:
|
|
if i in paired or j in paired:
|
|
continue
|
|
brightness_i = palette_np[i].sum()
|
|
brightness_j = palette_np[j].sum()
|
|
if brightness_i >= brightness_j:
|
|
auto_pairs.append((i, j))
|
|
else:
|
|
auto_pairs.append((j, i))
|
|
paired.add(i)
|
|
paired.add(j)
|
|
return auto_pairs
|
|
|
|
|
|
NODE_CLASS_MAPPINGS = {
|
|
"ColorshiftColor": ColorshiftColorNode,
|
|
"CsCPaletteEditor": PaletteEditorNode,
|
|
"CsCFill": CsCFill,
|
|
}
|
|
|
|
NODE_DISPLAY_NAME_MAPPINGS = {
|
|
"ColorshiftColor": "ColorshiftColor",
|
|
"CsCPaletteEditor": "CsCPaletteEditor",
|
|
"CsCFill": "CsCFill",
|
|
}
|