ls -la /home/daytona/workspace/ 2>/dev/null || mkdir -p /home/daytona/workspace/image-edit && echo "ready"

mkdir -p /home/daytona/workspace/image-edit && curl -s -o /home/daytona/workspace/image-edit/original.jpg "https://d361swluvexzqs.cloudfront.net/chats-attachments/3cae4348-e984-495b-9c1c-99d1a124b917_0D687A1E-19F3-1780555803104-jv0mm3.jpeg" && echo "Downloaded" && python3 -c "from PIL import Image; img = Image.open('/home/daytona/workspace/image-edit/original.jpg'); print(img.size)"

python3 << 'EOF' from PIL import Image, ImageFilter import numpy as np img = Image.open('/home/daytona/workspace/image-edit/original.jpg') w, h = img.size print(f"Size: {w}x{h}") # The male is roughly on the left 55% of the image # The female is roughly on the right half # Let's crop to just show the female, starting around x=480 # and fill the left side with extended background img_arr = np.array(img) # Crop to show only female - from roughly x=480 to end female_crop = img.crop((480, 0, w, h)) female_crop.save('/home/daytona/workspace/image-edit/female_only_crop.jpg', quality=95) print("Saved cropped version") print(f"Cropped size: {female_crop.size}") EOF

python3 << 'EOF' from PIL import Image, ImageFilter, ImageDraw import numpy as np img = Image.open('/home/daytona/workspace/image-edit/original.jpg') w, h = img.size img_arr = np.array(img) # Strategy: # 1. Keep the right portion (female side) intact # 2. For the left portion where the male was, reconstruct background # by sampling and extending the background from around the male # The male occupies roughly x=0 to x=580, y=130 to y=1050 # We'll create a new image same size but replace the male region with background # Sample background from the upper area (above heads) and sides # Background region: top strip y=0 to y=130, and far right column # Create output image out = img_arr.copy() # The background behind both people - sample from top area # Top background: rows 0-130, full width - this has the decorative lights top_bg = img_arr[0:130, :, :] # For the male region, we need to fill with the blurred/extended background # The background visible on far left edge and between/around them # Approach: use the region just to the right of male (background between them) # visible at roughly x=560-620 which is background between the two people # Also sample background from top-left corner # Let's do a smarter approach: # 1. Mask the male (left person) region # 2. Fill with background sampled from surrounding areas # Male bounding box (approximate) male_x1, male_y1 = 0, 130 male_x2, male_y2 = 610, 1060 # Create a filled version by: # - Taking the background column just to the right of male head (~x=610-650) # - Tiling/extending it leftward # - Blending with top background # Sample the background strip between the two people (x=590-650) bg_strip = img_arr[:, 590:660, :] # background visible between them # Also sample far left background visible above the male shoulder (x=0-50, y=0-180) # and the background in the upper left # Fill the male region result = img_arr.copy() # For each pixel in male region, blend background # Simple approach: extend the background strip strip_w = bg_strip.shape[1] for x in range(male_x2): # Use corresponding strip pixel, or mirror from background strip src_x = min(x % strip_w, strip_w - 1) # Blend more from right side for smoother transition blend = min(1.0, (male_x2 - x) / 200.0) # fade at right edge for y in range(male_y1, male_y2): bg_pixel = bg_strip[y, src_x, :] orig_pixel = img_arr[y, x, :] # In the core male body area, use background if x < male_x2 - 80: result[y, x, :] = bg_pixel else: # Blend zone near female t = (male_x2 - x) / 80.0 result[y, x, :] = (bg_pixel * t + orig_pixel * (1 - t)).astype(np.uint8) # Now apply blur to the filled region to make it look more natural result_img = Image.fromarray(result) # Blur just the replaced region mask = Image.new('L', (w, h), 0) draw = ImageDraw.Draw(mask) draw.rectangle([male_x1, male_y1, male_x2, male_y2], fill=255) blurred = result_img.filter(ImageFilter.GaussianBlur(radius=8)) result_img = Image.composite(blurred, result_img, mask) # Now crop to show only the female side cleanly # Crop from x=560 to end to remove all male traces final = result_img.crop((560, 0, w, h)) final.save('/home/daytona/workspace/image-edit/female_removed_male.jpg', quality=95) print(f"Saved. Final size: {final.size}") EOF

python3 << 'EOF' # Better approach: use OpenCV inpainting for more natural result import cv2 import numpy as np from PIL import Image img = cv2.imread('/home/daytona/workspace/image-edit/original.jpg') h, w = img.shape[:2] print(f"Image: {w}x{h}") # Create mask for the male (left person) # Male occupies roughly left half: x=0..590, body from y=130 to bottom mask = np.zeros((h, w), dtype=np.uint8) # Head area of male cv2.ellipse(mask, (255, 380), (165, 200), 0, 0, 360, 255, -1) # Body/torso of male cv2.rectangle(mask, (0, 520), (600, 1100), 255, -1) # Arm area cv2.rectangle(mask, (0, 400), (620, 580), 255, -1) # Left side background should not be masked (above male) # Keep the top part and right portion clear mask[:130, :] = 0 # top background - keep mask[:, 620:] = 0 # right side (female) - keep # Save mask for inspection cv2.imwrite('/home/daytona/workspace/image-edit/mask.jpg', mask) # Apply inpainting result = cv2.inpaint(img, mask, inpaintRadius=15, flags=cv2.INPAINT_TELEA) cv2.imwrite('/home/daytona/workspace/image-edit/inpainted_full.jpg', result) # Now crop to show only female side (right portion) # female is from about x=530 onwards female_region = result[:, 530:, :] cv2.imwrite('/home/daytona/workspace/image-edit/result_female_only.jpg', female_region, [cv2.IMWRITE_JPEG_QUALITY, 95]) print(f"Inpainting done. Female crop size: {female_region.shape[1]}x{female_region.shape[0]}") EOF

python3 << 'EOF' # Final polished version - better mask and crop import cv2 import numpy as np img = cv2.imread('/home/daytona/workspace/image-edit/original.jpg') h, w = img.shape[:2] # More precise mask for the male person mask = np.zeros((h, w), dtype=np.uint8) # Male head (centered around x=250, y=370) cv2.ellipse(mask, (250, 370), (175, 215), 0, 0, 360, 255, -1) # Neck cv2.rectangle(mask, (175, 560), (340, 650), 255, -1) # Shoulders and upper body cv2.rectangle(mask, (0, 580), (620, 720), 255, -1) # Lower body / torso cv2.rectangle(mask, (0, 700), (570, 1100), 255, -1) # Left arm cv2.rectangle(mask, (0, 600), (200, 1100), 255, -1) # Don't mask the table area at bottom mask[1100:, :] = 0 # Don't mask the top background mask[:140, :] = 0 # Don't mask female side mask[:, 600:] = 0 # Dilate mask slightly for cleaner edges kernel = np.ones((15, 15), np.uint8) mask = cv2.dilate(mask, kernel, iterations=1) mask[:, 600:] = 0 # re-zero female side after dilation # Save mask cv2.imwrite('/home/daytona/workspace/image-edit/mask_v2.jpg', mask) # Inpaint result = cv2.inpaint(img, mask, inpaintRadius=20, flags=cv2.INPAINT_TELEA) # Crop: keep from x=520 to end (female fully visible) final = result[:, 510:] cv2.imwrite('/home/daytona/workspace/image-edit/final_female.jpg', final, [cv2.IMWRITE_JPEG_QUALITY, 96]) print(f"Done. Output: {final.shape[1]}x{final.shape[0]}") EOF