Spaces:
Running
on
Zero
Running
on
Zero
| """ | |
| author: Chris Freilich | |
| description: This extension provides a blend modes node with 30 blend modes. | |
| """ | |
| from PIL import Image | |
| import numpy as np | |
| import torch | |
| import torch.nn.functional as F | |
| from colorsys import rgb_to_hsv | |
| from blend_modes import difference, normal, screen, soft_light, lighten_only, dodge, \ | |
| addition, darken_only, multiply, hard_light, \ | |
| grain_extract, grain_merge, divide, overlay | |
| def dissolve(backdrop, source, opacity): | |
| # Normalize the RGB and alpha values to 0-1 | |
| backdrop_norm = backdrop[:, :, :3] / 255 | |
| source_norm = source[:, :, :3] / 255 | |
| source_alpha_norm = source[:, :, 3] / 255 | |
| # Calculate the transparency of each pixel in the source image | |
| transparency = opacity * source_alpha_norm | |
| # Generate a random matrix with the same shape as the source image | |
| random_matrix = np.random.random(source.shape[:2]) | |
| # Create a mask where the random values are less than the transparency | |
| mask = random_matrix < transparency | |
| # Use the mask to select pixels from the source or backdrop | |
| blend = np.where(mask[..., None], source_norm, backdrop_norm) | |
| # Apply the alpha channel of the source image to the blended image | |
| new_rgb = (1 - source_alpha_norm[..., None]) * backdrop_norm + source_alpha_norm[..., None] * blend | |
| # Ensure the RGB values are within the valid range | |
| new_rgb = np.clip(new_rgb, 0, 1) | |
| # Convert the RGB values back to 0-255 | |
| new_rgb = new_rgb * 255 | |
| # Calculate the new alpha value by taking the maximum of the backdrop and source alpha channels | |
| new_alpha = np.maximum(backdrop[:, :, 3], source[:, :, 3]) | |
| # Create a new RGBA image with the calculated RGB and alpha values | |
| result = np.dstack((new_rgb, new_alpha)) | |
| return result | |
| def rgb_to_hsv_via_torch(rgb_numpy: np.ndarray, device=None) -> torch.Tensor: | |
| """ | |
| Convert an RGB image to HSV. | |
| :param rgb: A tensor of shape (3, H, W) where the three channels correspond to R, G, B. | |
| The values should be in the range [0, 1]. | |
| :return: A tensor of shape (3, H, W) where the three channels correspond to H, S, V. | |
| The hue (H) will be in the range [0, 1], while S and V will be in the range [0, 1]. | |
| """ | |
| if device is None: | |
| device = torch.device("cuda" if torch.cuda.is_available() else "cpu") | |
| rgb = torch.from_numpy(rgb_numpy).float().permute(2, 0, 1).to(device) | |
| r, g, b = rgb[0], rgb[1], rgb[2] | |
| max_val, _ = torch.max(rgb, dim=0) | |
| min_val, _ = torch.min(rgb, dim=0) | |
| delta = max_val - min_val | |
| h = torch.zeros_like(max_val) | |
| s = torch.zeros_like(max_val) | |
| v = max_val | |
| # calc hue... avoid div by zero (by masking the delta) | |
| mask = delta != 0 | |
| r_eq_max = (r == max_val) & mask | |
| g_eq_max = (g == max_val) & mask | |
| b_eq_max = (b == max_val) & mask | |
| h[r_eq_max] = (g[r_eq_max] - b[r_eq_max]) / delta[r_eq_max] % 6 | |
| h[g_eq_max] = (b[g_eq_max] - r[g_eq_max]) / delta[g_eq_max] + 2.0 | |
| h[b_eq_max] = (r[b_eq_max] - g[b_eq_max]) / delta[b_eq_max] + 4.0 | |
| h = (h / 6.0) % 1.0 | |
| # calc saturation | |
| s[max_val != 0] = delta[max_val != 0] / max_val[max_val != 0] | |
| hsv = torch.stack([h, s, v], dim=0) | |
| hsv_numpy = hsv.permute(1, 2, 0).cpu().numpy() | |
| return hsv_numpy | |
| def hsv_to_rgb_via_torch(hsv_numpy: np.ndarray, device=None) -> torch.Tensor: | |
| """ | |
| Convert an HSV image to RGB. | |
| :param hsv: A tensor of shape (3, H, W) where the three channels correspond to H, S, V. | |
| The H channel values should be in the range [0, 1], while S and V will be in the range [0, 1]. | |
| :return: A tensor of shape (3, H, W) where the three channels correspond to R, G, B. | |
| The RGB values will be in the range [0, 1]. | |
| """ | |
| if device is None: | |
| device = torch.device("cuda" if torch.cuda.is_available() else "cpu") | |
| hsv = torch.from_numpy(hsv_numpy).float().permute(2, 0, 1).to(device) | |
| h, s, v = hsv[0], hsv[1], hsv[2] | |
| c = v * s # chroma | |
| x = c * (1 - torch.abs((h * 6) % 2 - 1)) | |
| m = v - c # match value | |
| z = torch.zeros_like(h) | |
| rgb = torch.zeros_like(hsv) | |
| # define conditions for different hue ranges | |
| h_cond = [ | |
| (h < 1/6, torch.stack([c, x, z], dim=0)), | |
| ((1/6 <= h) & (h < 2/6), torch.stack([x, c, z], dim=0)), | |
| ((2/6 <= h) & (h < 3/6), torch.stack([z, c, x], dim=0)), | |
| ((3/6 <= h) & (h < 4/6), torch.stack([z, x, c], dim=0)), | |
| ((4/6 <= h) & (h < 5/6), torch.stack([x, z, c], dim=0)), | |
| (h >= 5/6, torch.stack([c, z, x], dim=0)), | |
| ] | |
| # conditionally set RGB values based on the hue range | |
| for cond, result in h_cond: | |
| rgb[:, cond] = result[:, cond] | |
| # add match value to convert to final RGB values | |
| rgb = rgb + m | |
| rgb_numpy = rgb.permute(1, 2, 0).cpu().numpy() | |
| return rgb_numpy | |
| def hsv(backdrop, source, opacity, channel): | |
| # Convert RGBA to RGB, normalized | |
| backdrop_rgb = backdrop[:, :, :3] / 255.0 | |
| source_rgb = source[:, :, :3] / 255.0 | |
| source_alpha = source[:, :, 3] / 255.0 | |
| # Convert RGB to HSV | |
| backdrop_hsv = rgb_to_hsv_via_torch(backdrop_rgb) | |
| source_hsv = rgb_to_hsv_via_torch(source_rgb) | |
| # Combine HSV values | |
| new_hsv = backdrop_hsv.copy() | |
| # Determine which channel to operate on | |
| if channel == "saturation": | |
| new_hsv[:, :, 1] = (1 - opacity * source_alpha) * backdrop_hsv[:, :, 1] + opacity * source_alpha * source_hsv[:, :, 1] | |
| elif channel == "luminance": | |
| new_hsv[:, :, 2] = (1 - opacity * source_alpha) * backdrop_hsv[:, :, 2] + opacity * source_alpha * source_hsv[:, :, 2] | |
| elif channel == "hue": | |
| new_hsv[:, :, 0] = (1 - opacity * source_alpha) * backdrop_hsv[:, :, 0] + opacity * source_alpha * source_hsv[:, :, 0] | |
| elif channel == "color": | |
| new_hsv[:, :, :2] = (1 - opacity * source_alpha[..., None]) * backdrop_hsv[:, :, :2] + opacity * source_alpha[..., None] * source_hsv[:, :, :2] | |
| # Convert HSV back to RGB | |
| new_rgb = hsv_to_rgb_via_torch(new_hsv) | |
| # Apply the alpha channel of the source image to the new RGB image | |
| new_rgb = (1 - source_alpha[..., None]) * backdrop_rgb + source_alpha[..., None] * new_rgb | |
| # Ensure the RGB values are within the valid range | |
| new_rgb = np.clip(new_rgb, 0, 1) | |
| # Convert RGB back to RGBA and scale to 0-255 range | |
| new_rgba = np.dstack((new_rgb * 255, backdrop[:, :, 3])) | |
| return new_rgba.astype(np.uint8) | |
| def saturation(backdrop, source, opacity): | |
| return hsv(backdrop, source, opacity, "saturation") | |
| def luminance(backdrop, source, opacity): | |
| return hsv(backdrop, source, opacity, "luminance") | |
| def hue(backdrop, source, opacity): | |
| return hsv(backdrop, source, opacity, "hue") | |
| def color(backdrop, source, opacity): | |
| return hsv(backdrop, source, opacity, "color") | |
| def darker_lighter_color(backdrop, source, opacity, type): | |
| # Normalize the RGB and alpha values to 0-1 | |
| backdrop_norm = backdrop[:, :, :3] / 255 | |
| source_norm = source[:, :, :3] / 255 | |
| source_alpha_norm = source[:, :, 3] / 255 | |
| # Convert RGB to HSV | |
| backdrop_hsv = np.array([rgb_to_hsv(*rgb) for row in backdrop_norm for rgb in row]).reshape(backdrop.shape[:2] + (3,)) | |
| source_hsv = np.array([rgb_to_hsv(*rgb) for row in source_norm for rgb in row]).reshape(source.shape[:2] + (3,)) | |
| # Create a mask where the value (brightness) of the source image is less than the value of the backdrop image | |
| if type == "dark": | |
| mask = source_hsv[:, :, 2] < backdrop_hsv[:, :, 2] | |
| else: | |
| mask = source_hsv[:, :, 2] > backdrop_hsv[:, :, 2] | |
| # Use the mask to select pixels from the source or backdrop | |
| blend = np.where(mask[..., None], source_norm, backdrop_norm) | |
| # Apply the alpha channel of the source image to the blended image | |
| new_rgb = (1 - source_alpha_norm[..., None] * opacity) * backdrop_norm + source_alpha_norm[..., None] * opacity * blend | |
| # Ensure the RGB values are within the valid range | |
| new_rgb = np.clip(new_rgb, 0, 1) | |
| # Convert the RGB values back to 0-255 | |
| new_rgb = new_rgb * 255 | |
| # Calculate the new alpha value by taking the maximum of the backdrop and source alpha channels | |
| new_alpha = np.maximum(backdrop[:, :, 3], source[:, :, 3]) | |
| # Create a new RGBA image with the calculated RGB and alpha values | |
| result = np.dstack((new_rgb, new_alpha)) | |
| return result | |
| def darker_color(backdrop, source, opacity): | |
| return darker_lighter_color(backdrop, source, opacity, "dark") | |
| def lighter_color(backdrop, source, opacity): | |
| return darker_lighter_color(backdrop, source, opacity, "light") | |
| def simple_mode(backdrop, source, opacity, mode): | |
| # Normalize the RGB and alpha values to 0-1 | |
| backdrop_norm = backdrop[:, :, :3] / 255 | |
| source_norm = source[:, :, :3] / 255 | |
| source_alpha_norm = source[:, :, 3:4] / 255 | |
| # Calculate the blend without any transparency considerations | |
| if mode == "linear_burn": | |
| blend = backdrop_norm + source_norm - 1 | |
| elif mode == "linear_light": | |
| blend = backdrop_norm + (2 * source_norm) - 1 | |
| elif mode == "color_dodge": | |
| blend = backdrop_norm / (1 - source_norm) | |
| blend = np.clip(blend, 0, 1) | |
| elif mode == "color_burn": | |
| blend = 1 - ((1 - backdrop_norm) / source_norm) | |
| blend = np.clip(blend, 0, 1) | |
| elif mode == "exclusion": | |
| blend = backdrop_norm + source_norm - (2 * backdrop_norm * source_norm) | |
| elif mode == "subtract": | |
| blend = backdrop_norm - source_norm | |
| elif mode == "vivid_light": | |
| blend = np.where(source_norm <= 0.5, backdrop_norm / (1 - 2 * source_norm), 1 - (1 -backdrop_norm) / (2 * source_norm - 0.5) ) | |
| blend = np.clip(blend, 0, 1) | |
| elif mode == "pin_light": | |
| blend = np.where(source_norm <= 0.5, np.minimum(backdrop_norm, 2 * source_norm), np.maximum(backdrop_norm, 2 * (source_norm - 0.5))) | |
| elif mode == "hard_mix": | |
| blend = simple_mode(backdrop, source, opacity, "linear_light") | |
| blend = np.round(blend[:, :, :3] / 255) | |
| # Apply the blended layer back onto the backdrop layer while utilizing the alpha channel and opacity information | |
| new_rgb = (1 - source_alpha_norm * opacity) * backdrop_norm + source_alpha_norm * opacity * blend | |
| # Ensure the RGB values are within the valid range | |
| new_rgb = np.clip(new_rgb, 0, 1) | |
| # Convert the RGB values back to 0-255 | |
| new_rgb = new_rgb * 255 | |
| # Calculate the new alpha value by taking the maximum of the backdrop and source alpha channels | |
| new_alpha = np.maximum(backdrop[:, :, 3], source[:, :, 3]) | |
| # Create a new RGBA image with the calculated RGB and alpha values | |
| result = np.dstack((new_rgb, new_alpha)) | |
| return result | |
| def linear_light(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "linear_light") | |
| def vivid_light(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "vivid_light") | |
| def pin_light(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "pin_light") | |
| def hard_mix(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "hard_mix") | |
| def linear_burn(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "linear_burn") | |
| def color_dodge(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "color_dodge") | |
| def color_burn(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "color_burn") | |
| def exclusion(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "exclusion") | |
| def subtract(backdrop, source, opacity): | |
| return simple_mode(backdrop, source, opacity, "subtract") | |
| BLEND_MODES = { | |
| "normal": normal, | |
| "dissolve": dissolve, | |
| "darken": darken_only, | |
| "multiply": multiply, | |
| "color burn": color_burn, | |
| "linear burn": linear_burn, | |
| "darker color": darker_color, | |
| "lighten": lighten_only, | |
| "screen": screen, | |
| "color dodge": color_dodge, | |
| "linear dodge(add)": addition, | |
| "lighter color": lighter_color, | |
| "dodge": dodge, | |
| "overlay": overlay, | |
| "soft light": soft_light, | |
| "hard light": hard_light, | |
| "vivid light": vivid_light, | |
| "linear light": linear_light, | |
| "pin light": pin_light, | |
| "hard mix": hard_mix, | |
| "difference": difference, | |
| "exclusion": exclusion, | |
| "subtract": subtract, | |
| "divide": divide, | |
| "hue": hue, | |
| "saturation": saturation, | |
| "color": color, | |
| "luminosity": luminance, | |
| "grain extract": grain_extract, | |
| "grain merge": grain_merge | |
| } | |