When the AI returns low-confidence results or completely fails to identify an item, there's almost always a fixable cause. Work through these steps.
Most Common Cause: Photo Quality
Take a new photo before anything else. A better photo fixes 80% of AI failures.
Check for:
- Blurriness — the image must be sharp enough to read markings. Tap to focus before shooting.
- Bad lighting — overexposed (too bright) or underexposed (too dark). Use indirect natural light near a window.
- Angle too steep — shooting from above instead of at eye level makes markings hard to read.
- Background clutter — a busy layout background makes it hard for the AI to isolate the model. Use a plain white or gray card behind the model.
See Photo Tips for a complete guide.
The Item Has No Markings
Some items legitimately have no prototype markings:
- Undecorated brass models — no road name, no paint
- Vintage tinplate toys — generic markings that don't match prototype data
- Kit-bashed or scratch-built models — no manufacturer data to match
For these, use Manual Entry. The AI skips to manual entry automatically if it detects no recognizable markings.
Obscure Prototype Road
The AI has more training data for major US Class I railroads (UP, BNSF, CSX, NS, CP, CN) than for shortlines, narrow-gauge railroads, or foreign railroads.
If you're modeling:
- A shortline (e.g., Livermore & Amador Valley)
- A narrow-gauge railroad (e.g., D&RGW Silverton Branch)
- A foreign railroad (e.g., DB, SNCF, BR)
The AI may not recognize the specific livery. Enter the details manually. Your corrections help train the model for these under-represented prototypes.
Scale Too Small
Z-scale models have markings smaller than 1mm in real life. Even excellent photos sometimes lack the resolution for text recognition at Z scale. HO and N scale work best for AI recognition.
Wrong Photo Type
The AI is designed for side-on model photos, not:
- Packaging / box photos (the box art, not the model)
- Catalog scans
- Scale-drawing reference photos
Always photograph the actual model, not reference materials.
Try Re-uploading the Same Photo
Occasionally an upload fails partway through and the AI receives a corrupted image. Delete the photo from the item and re-upload it from the original file.
Report Persistent Failures
If you've taken a good photo and the AI still fails on an item that should be recognizable:
- Save the item with correct manual data
- Go to the Photos tab → click Report AI Error on the photo you used
- Describe what it should have identified
This flags the case for engineering review and helps improve the model for similar equipment.
Next Steps
- Photo Tips — the complete photography guide
- Improving AI Accuracy — contribute corrections
- How AI Recognition Works — understand the limitations