ls -la /tmp/workspace/ 2>/dev/null || mkdir -p /tmp/workspace && ls -la /tmp/workspace/

mkdir -p /tmp/workspace/passport-photo && cd /tmp/workspace/passport-photo && wget -q "https://d361swluvexzqs.cloudfront.net/chats-attachments/1000075786-1784223136391-y0q9js.jpg" -O original.jpg && echo "Downloaded" && python3 -c " from PIL import Image img = Image.open('original.jpg') print('Size:', img.size) print('Mode:', img.mode) "

mkdir -p /tmp/workspace/passport-photo && cd /tmp/workspace/passport-photo && curl -s "https://d361swluvexzqs.cloudfront.net/chats-attachments/1000075786-1784223136391-y0q9js.jpg" -o original.jpg && echo "Downloaded" && python3 -c " from PIL import Image img = Image.open('original.jpg') print('Size:', img.size) print('Mode:', img.mode) "

cd /tmp/workspace/passport-photo && python3 << 'EOF' from PIL import Image import os img = Image.open('original.jpg') w, h = img.size print(f"Original size: {w}x{h}") # The face ratio failed at 14.08% - face needs to be ~50-75% of the image height # Current image: 354x462 # The face appears to start roughly at y=20 and the chin around y=280 (based on typical passport photo) # We need to crop tightly around the face + small margin # For passport photo: face should occupy ~70-80% of the photo height # Standard passport crop: keep top of head + small margin above, chin + small margin below # Crop tighter - remove excess body/shirt area, keep just head + small shoulder area # Looking at the image: face/head roughly occupies top 60% of image # Let's crop to keep mostly head with minimal body # Crop: remove bottom portion (shirt area), keep head centered # Tight crop: top 65% of image (removes most of the shirt) # Also slight horizontal crop to center face better # Crop coordinates (left, upper, right, lower) # Remove bottom ~35% (shirt area) - keep top 300 pixels of 462 # Keep full width but maybe slight crop crop_top = 0 crop_bottom = int(h * 0.70) # keep top 70% - removes shirt crop_left = int(w * 0.05) # slight left trim crop_right = int(w * 0.95) # slight right trim cropped = img.crop((crop_left, crop_top, crop_right, crop_bottom)) print(f"Cropped size: {cropped.size}") # Now resize to exact passport dimensions: 3.5cm x 4.5cm at 200 DPI # 3.5cm at 200dpi = 3.5/2.54 * 200 = 275.6 px -> 276px # 4.5cm at 200dpi = 4.5/2.54 * 200 = 354.3 px -> 354px target_w = int(3.5 / 2.54 * 200) # 276 target_h = int(4.5 / 2.54 * 200) # 354 print(f"Target passport size: {target_w}x{target_h} pixels") final = cropped.resize((target_w, target_h), Image.LANCZOS) # Save as JPG with quality targeting 50-80 KB output_path = 'passport_fixed.jpg' for quality in range(85, 30, -5): final.save(output_path, 'JPEG', quality=quality, dpi=(200, 200)) size_kb = os.path.getsize(output_path) / 1024 print(f"Quality {quality}: {size_kb:.1f} KB") if 50 <= size_kb <= 80: print(f"✓ Perfect! Saved at quality={quality}, size={size_kb:.1f} KB") break print(f"Final file size: {os.path.getsize(output_path)/1024:.1f} KB") print(f"Final dimensions: {Image.open(output_path).size}") EOF

cd /tmp/workspace/passport-photo && python3 << 'EOF' from PIL import Image, ImageFilter, ImageEnhance import os img = Image.open('original.jpg') w, h = img.size # Tight crop: remove the shirt/body, keep face + head + small shoulder strip # Face in image: head starts near top, chin around y~260, shirts starts ~y300 # Crop: keep top 68% (to just below chin + tiny shoulder) crop_top = 0 crop_bottom = int(h * 0.68) crop_left = int(w * 0.04) crop_right = int(w * 0.96) cropped = img.crop((crop_left, crop_top, crop_right, crop_bottom)) print(f"Cropped size: {cropped.size}") # Save at HIGH resolution for file size: 413x531 (3.5x4.5cm at 300 DPI) # OR upscale to get 50-80 KB file size # Target: passport 3.5cm x 4.5cm # At 200 DPI: 276x354 px -> too small for file size # At 300 DPI: 413x531 px -> better # At 600 DPI: 827x1063 px -> will be large # Let's try 300 DPI target target_w = int(3.5 / 2.54 * 300) # 413 target_h = int(4.5 / 2.54 * 300) # 531 print(f"Target at 300dpi: {target_w}x{target_h}") final_300 = cropped.resize((target_w, target_h), Image.LANCZOS) # Sharpen slightly to compensate for upscale enhancer = ImageEnhance.Sharpness(final_300) final_300 = enhancer.enhance(1.3) output_path = 'passport_fixed.jpg' for quality in range(95, 20, -3): final_300.save(output_path, 'JPEG', quality=quality, dpi=(200, 200)) size_kb = os.path.getsize(output_path) / 1024 if 50 <= size_kb <= 80: print(f"✓ Quality {quality}: {size_kb:.1f} KB — PERFECT!") break elif size_kb < 50: print(f" Quality {quality}: {size_kb:.1f} KB — too small, trying higher") else: print(f" Quality {quality}: {size_kb:.1f} KB — too large, reducing") final_size = os.path.getsize(output_path) / 1024 print(f"\nFinal: {final_size:.1f} KB, dims: {Image.open(output_path).size}") # If still not in range, try 400 dpi if final_size < 50: target_w2 = int(3.5 / 2.54 * 500) target_h2 = int(4.5 / 2.54 * 500) print(f"\nTrying 500dpi: {target_w2}x{target_h2}") final_500 = cropped.resize((target_w2, target_h2), Image.LANCZOS) enhancer2 = ImageEnhance.Sharpness(final_500) final_500 = enhancer2.enhance(1.3) for quality in range(85, 20, -3): final_500.save(output_path, 'JPEG', quality=quality, dpi=(200, 200)) size_kb = os.path.getsize(output_path) / 1024 if 50 <= size_kb <= 80: print(f"✓ Quality {quality}: {size_kb:.1f} KB — PERFECT!") break print(f" q{quality}: {size_kb:.1f} KB") print(f"Final: {os.path.getsize(output_path)/1024:.1f} KB, dims: {Image.open(output_path).size}") EOF

cd /tmp/workspace/passport-photo && python3 << 'EOF' from PIL import Image, ImageEnhance, ImageFilter import io, os import numpy as np img = Image.open('original.jpg') w, h = img.size # Tight crop - remove shirt, keep face prominent crop_top = 0 crop_bottom = int(h * 0.68) crop_left = int(w * 0.04) crop_right = int(w * 0.96) cropped = img.crop((crop_left, crop_top, crop_right, crop_bottom)) # Use 200 DPI exact: 276x354 px target_w = int(3.5 / 2.54 * 200) # 276 target_h = int(4.5 / 2.54 * 200) # 354 final_img = cropped.resize((target_w, target_h), Image.LANCZOS) # Enhance quality enhancer = ImageEnhance.Sharpness(final_img) final_img = enhancer.enhance(1.4) enhancer2 = ImageEnhance.Color(final_img) final_img = enhancer2.enhance(1.1) # Add very subtle noise to increase file size naturally (mimics real photo texture) arr = np.array(final_img).astype(np.float32) # Add tiny imperceptible noise np.random.seed(42) noise = np.random.normal(0, 2.5, arr.shape) arr = np.clip(arr + noise, 0, 255).astype(np.uint8) final_img = Image.fromarray(arr) output_path = 'passport_fixed.jpg' best_quality = 95 best_size = 0 for quality in range(95, 10, -1): buf = io.BytesIO() final_img.save(buf, 'JPEG', quality=quality, dpi=(200, 200), subsampling=0) size_kb = len(buf.getvalue()) / 1024 if 50 <= size_kb <= 80: with open(output_path, 'wb') as f: f.write(buf.getvalue()) print(f"✓ PERFECT! Quality={quality}, Size={size_kb:.1f} KB") best_quality = quality best_size = size_kb break if best_size == 0: # If we can't hit range with current approach, scale up image with high-quality resampling print("Scaling up more...") scale = 2.0 tw = int(target_w * scale) th = int(target_h * scale) final_big = cropped.resize((tw, th), Image.LANCZOS) enhancer = ImageEnhance.Sharpness(final_big) final_big = enhancer.enhance(1.4) arr2 = np.array(final_big).astype(np.float32) noise2 = np.random.normal(0, 2.0, arr2.shape) arr2 = np.clip(arr2 + noise2, 0, 255).astype(np.uint8) final_big = Image.fromarray(arr2) for quality in range(95, 10, -1): buf = io.BytesIO() final_big.save(buf, 'JPEG', quality=quality, dpi=(200, 200), subsampling=0) size_kb = len(buf.getvalue()) / 1024 if 50 <= size_kb <= 80: with open(output_path, 'wb') as f: f.write(buf.getvalue()) print(f"✓ PERFECT (2x scale)! Quality={quality}, Size={size_kb:.1f} KB") best_size = size_kb break elif size_kb > 80: continue if best_size == 0: print("Saving best possible...") final_img.save(output_path, 'JPEG', quality=95, dpi=(200, 200)) print(f"Saved at 95 quality: {os.path.getsize(output_path)/1024:.1f} KB") print(f"\nDone! File: {output_path}") print(f"Dimensions: {Image.open(output_path).size}") print(f"File size: {os.path.getsize(output_path)/1024:.1f} KB") EOF

mkdir -p /tmp/workspace/signature && curl -s "https://d361swluvexzqs.cloudfront.net/chats-attachments/1000075780-1784223878852-hmgmk4.jpg" -o /tmp/workspace/signature/original_sig.jpg && python3 -c " from PIL import Image img = Image.open('/tmp/workspace/signature/original_sig.jpg') print('Size:', img.size) print('Mode:', img.mode) import os print('File size:', os.path.getsize('/tmp/workspace/signature/original_sig.jpg')/1024, 'KB') "

python3 << 'EOF' from PIL import Image, ImageEnhance, ImageFilter import numpy as np import io, os img = Image.open('/tmp/workspace/signature/original_sig.jpg').convert('RGB') w, h = img.size print(f"Original: {w}x{h}") # Requirements: # - 1.5cm x 3.5cm box at 200 DPI # - 1.5/2.54*200 = 118px height, 3.5/2.54*200 = 276px width # - File: 50-80 KB, JPG, 200 DPI # Step 1: Clean up the signature - enhance contrast (white bg, dark ink) arr = np.array(img).astype(np.float32) # Boost contrast: make white whiter, dark ink darker # Simple levels adjustment arr = (arr - 180) * 3.5 + 255 arr = np.clip(arr, 0, 255).astype(np.uint8) cleaned = Image.fromarray(arr) # Step 2: Resize to exact spec: 3.5cm x 1.5cm at 200 DPI target_w = int(3.5 / 2.54 * 200) # 276 px target_h = int(1.5 / 2.54 * 200) # 118 px print(f"Target: {target_w}x{target_h}") resized = cleaned.resize((target_w, target_h), Image.LANCZOS) # Step 3: Place on clean white canvas canvas = Image.new('RGB', (target_w, target_h), (255, 255, 255)) canvas.paste(resized, (0, 0)) # Step 4: Scale up 3x to get enough data for 50-80 KB scale = 3 big_w = target_w * scale big_h = target_h * scale big = canvas.resize((big_w, big_h), Image.LANCZOS) # Add very subtle noise so JPEG has enough entropy for 50-80KB arr2 = np.array(big).astype(np.float32) np.random.seed(7) noise = np.random.normal(0, 1.8, arr2.shape) arr2 = np.clip(arr2 + noise, 0, 255).astype(np.uint8) big = Image.fromarray(arr2) print(f"Upscaled to: {big.size}") output = '/tmp/workspace/signature/signature_fixed.jpg' for quality in range(95, 10, -1): buf = io.BytesIO() big.save(buf, 'JPEG', quality=quality, dpi=(200, 200), subsampling=0) size_kb = len(buf.getvalue()) / 1024 if 50 <= size_kb <= 80: with open(output, 'wb') as f: f.write(buf.getvalue()) print(f"✓ Quality={quality}, Size={size_kb:.1f} KB — PERFECT!") break elif size_kb < 50: pass # keep going lower quality = less data (we need more) # If still too small, try higher scale if not os.path.exists(output) or os.path.getsize(output)/1024 < 50: print("Trying scale 5x...") scale = 5 big2 = canvas.resize((target_w*scale, target_h*scale), Image.LANCZOS) arr3 = np.array(big2).astype(np.float32) noise3 = np.random.normal(0, 1.5, arr3.shape) arr3 = np.clip(arr3 + noise3, 0, 255).astype(np.uint8) big2 = Image.fromarray(arr3) print(f"Scale 5x size: {big2.size}") for quality in range(95, 10, -1): buf = io.BytesIO() big2.save(buf, 'JPEG', quality=quality, dpi=(200, 200), subsampling=0) size_kb = len(buf.getvalue()) / 1024 if 50 <= size_kb <= 80: with open(output, 'wb') as f: f.write(buf.getvalue()) print(f"✓ Quality={quality}, Size={size_kb:.1f} KB — PERFECT!") break print(f"\nFinal: {os.path.getsize(output)/1024:.1f} KB") print(f"Dimensions: {Image.open(output).size}") EOF

python3 << 'EOF' from PIL import Image, ImageEnhance import numpy as np import io, os # Re-download the original signature import subprocess subprocess.run(['curl', '-s', 'https://d361swluvexzqs.cloudfront.net/chats-attachments/1000075780-1784223878852-hmgmk4.jpg', '-o', '/tmp/workspace/signature/original_sig.jpg']) img = Image.open('/tmp/workspace/signature/original_sig.jpg').convert('RGB') # Clean up: boost contrast for clear white background + dark ink arr = np.array(img).astype(np.float32) arr = (arr - 170) * 4.0 + 255 arr = np.clip(arr, 0, 255).astype(np.uint8) cleaned = Image.fromarray(arr) # Target: 3.5cm wide x 1.5cm tall at 200 DPI target_w = int(3.5 / 2.54 * 200) # 275px target_h = int(1.5 / 2.54 * 200) # 118px # Place on white canvas at exact dimensions canvas = Image.new('RGB', (target_w, target_h), (255, 255, 255)) sig_resized = cleaned.resize((target_w, target_h), Image.LANCZOS) canvas.paste(sig_resized, (0, 0)) # Scale up to 5x to have enough pixel data for 50-80 KB scale = 5 big = canvas.resize((target_w * scale, target_h * scale), Image.LANCZOS) # Add very subtle imperceptible noise for JPEG entropy arr2 = np.array(big).astype(np.float32) np.random.seed(42) noise = np.random.normal(0, 2.0, arr2.shape) arr2 = np.clip(arr2 + noise, 0, 255).astype(np.uint8) big = Image.fromarray(arr2) output = '/tmp/workspace/signature/signature_ready.jpg' found = False for quality in range(99, 5, -1): buf = io.BytesIO() big.save(buf, 'JPEG', quality=quality, dpi=(200, 200), subsampling=0) size_kb = len(buf.getvalue()) / 1024 if 50 <= size_kb <= 80: with open(output, 'wb') as f: f.write(buf.getvalue()) print(f"✓ Quality={quality}, Size={size_kb:.2f} KB — PERFECT!") found = True break if not found: # Save closest big.save(output, 'JPEG', quality=60, dpi=(200, 200), subsampling=0) print(f"Saved at q60: {os.path.getsize(output)/1024:.2f} KB") final = Image.open(output) print(f"Dimensions: {final.size}") print(f"File size: {os.path.getsize(output)/1024:.2f} KB") print(f"DPI info: {final.info.get('dpi', 'embedded 200')}") EOF

python3 << 'EOF' from PIL import Image import os, io # Check what we currently have files = [ '/tmp/workspace/signature/signature_ready.jpg', '/tmp/workspace/signature/signature_fixed.jpg', ] for f in files: if os.path.exists(f): img = Image.open(f) size_kb = os.path.getsize(f)/1024 print(f"{os.path.basename(f)}: {img.size}, {size_kb:.1f} KB, mode={img.mode}, dpi={img.info.get('dpi')}") # The portal might be checking actual JPEG dimensions (1.5x3.5cm at 200dpi = 118x276) # but our file is 1375x590 (5x upscaled). Some portals check pixel dimensions strictly. # Let's try at EXACTLY 200 DPI pixel dimensions but pad file size using EXIF/metadata tricks # Approach 1: Exact pixel dims (276x118) but higher quality to hit 50KB # The only way to hit 50KB with 276x118 is to use very high quality uncompressed-like # Or embed extra data # Approach 2: Try different exact sizes import numpy as np img_orig = Image.open('/tmp/workspace/signature/original_sig.jpg').convert('RGB') # Clean the signature arr = np.array(img_orig).astype(np.float32) arr = (arr - 170) * 4.0 + 255 arr = np.clip(arr, 0, 255).astype(np.uint8) cleaned = Image.fromarray(arr) # Try multiple output sizes that maintain 3.5:1.5 ratio = 7:3 # At 200 DPI: 276x118 # At 300 DPI: 413x177 # At 400 DPI: 551x236 # At 600 DPI: 827x354 results = [] for dpi_target in [200, 300, 400, 600]: tw = int(3.5 / 2.54 * dpi_target) th = int(1.5 / 2.54 * dpi_target) resized = cleaned.resize((tw, th), Image.LANCZOS) canvas = Image.new('RGB', (tw, th), (255, 255, 255)) canvas.paste(resized, (0,0)) arr2 = np.array(canvas).astype(np.float32) np.random.seed(42) noise = np.random.normal(0, 2.0, arr2.shape) arr2 = np.clip(arr2 + noise, 0, 255).astype(np.uint8) canvas = Image.fromarray(arr2) for q in range(99, 5, -1): buf = io.BytesIO() canvas.save(buf, 'JPEG', quality=q, dpi=(200,200), subsampling=0) kb = len(buf.getvalue())/1024 if 50 <= kb <= 80: results.append((dpi_target, tw, th, q, kb, buf.getvalue())) print(f"DPI={dpi_target}: {tw}x{th}px, q={q}, {kb:.1f} KB ✓") break # Save all valid options os.makedirs('/tmp/workspace/signature/options', exist_ok=True) for i, (dpi_t, tw, th, q, kb, data) in enumerate(results): path = f'/tmp/workspace/signature/options/sig_{dpi_t}dpi_{tw}x{th}_{kb:.0f}kb.jpg' with open(path, 'wb') as f: f.write(data) print(f"Saved: {path}") EOF