I just shipped a new feature for kleinanz.weedygallery.de that makes your photos look better automatically – intelligent auto-cropping for clothing items.
✨ What's the problem?
When people photograph clothes for resale, they waste precious time:
Manual cropping: Users have to crop out distracting backgrounds, cluttered rooms, faces/bodies manually
Messy photos: Extra background (walls, floors, clutter) distracts from the actual product
Tedious editing: Every photo needs cropping work – multiply that by a handful of images, and you're stuck in photo-editing hell 😓
Buyers need to see the clothes, not your living room. But editing shouldn't be your job – you want to sell fast, not play photographer.
Note: This cuts down the image frame to focus on the clothing – it's smart cropping, not background removal or color-based edits. No fancy AI filters, just efficient framing. 📸
The Smart Process
Here's the magic behind the feature:
1️⃣ Upload your photo User uploads an image of a clothing item (shirt, pants, dress, jacket, etc.)
2️⃣ ML detects the clothing My system uses a fashion-specific YOLO object detection model (trained on clothing) to find all clothing items in the image. It returns a confidence score for each detection.
3️⃣ Filter low-confidence detections If the model is unsure (below certain confidence), it ignores that detection, so its another product than clothing.
4️⃣ Check if already a close-up This is the smart part: the system measures how much of the image the largest detected clothing item covers.
Already 70%+ of the image? → It's already a close-up, no crop needed ✓
Less than 70%? → Crop it!
5️⃣ Smart cropping with margin The system locks onto the main (largest) clothing item and crops to it – but adds a smart margin around it so you still see the context and overall fit. This isn't a harsh, centerless crop.
6️⃣ Return the result The cropped image is now focused on the clothing, background is gone, and your product stands out. On error? No problem – the original image is returned, fail-safe.
Nerd specs
YOLO Model: Fashion-specific YOLOv8 detection (pure Python via ONNX Runtime – works flawlessly on Streamlit Cloud)
Threshold Logic: Confidence filtering + close-up detection prevents over-cropping
Safe Fallback: If anything goes wrong, the original image is used
Pure Python: No heavy dependencies, keeps the app lightweight
🎯 The result
Skip the cropping. Sell faster. ⚡
Users upload a messy photo → AI crops it smart → the product shines → better small ads in seconds.
That's it. No photo editing tedium, no wasted clicks. Just upload, generate your ad text, done. ✅
🍨 Support
Building AI-powered tools that run smoothly on the cloud takes effort. If you love this feature and want to help keep kleinanz running (and keep my ice cream supply stocked 🍦), your support means everything.
Thanks for being part of this journey! 🙌
Try it now: kleinanz.streamlit.io
Buy me a coffee: buymeacoffee.com/butburg