RailScanPro's AI Vision reads photos the way an experienced collector reads a model — looking for markings, livery patterns, shape details, and prototype clues. Here's how it works.
What the AI Reads from Your Photo
When you upload a photo, the AI analyzes several layers of information simultaneously:
Text and markings — Road names, reporting marks, road numbers, builder's plate text, weight and capacity stencils. The AI can read text printed on a model even when it's very small.
Livery and color patterns — The color blocking, stripe patterns, herald shapes, and font styles that distinguish, for example, a classic Electro-Motive two-tone gray from a 1980s Norfolk Southern black scheme.
Shape and silhouette — The overall profile: hood shape, truck style, number of axles, cab style, fuel tank profile. These are strong signals for identifying locomotive classes even when markings are faded.
Era indicators — Prototype details that place equipment in a historical period: headlight style (single sealed-beam vs. oscillating), plow vs. no plow, cab signal antennas, classification lights.
How Confident Is the AI?
Each identification comes with a confidence score shown as a percentage. When you see the AI's suggestions:
- 90%+ confidence — almost certainly correct; the AI has strong evidence
- 70–89% — very likely correct; review and confirm
- 50–69% — plausible match; check that road name and model align with your model
- Below 50% — the AI is making a best guess; manual review recommended
You're always in control — the AI's suggestions are pre-fills, not locked values. Correct anything that's wrong before saving.
What the AI Is Good At
- Major US prototype railroads in HO and N scale — where the training data is deepest
- Modern locomotives with clear markings — GP38-2, SD40-2, AC4400CW, ES44AC, and similar modern power
- Popular steam locomotive classes — USRA Light Mikado, Pennsy K4, Santa Fe 2-10-4
- Freight car stenciling — AAR reporting marks, COTS-stenciled hoppers, refrigerator car heralds
Where the AI Is Less Certain
- Heavily weathered models where markings are obscured
- Brass undecorated models with no paint or lettering
- Small scale details on Z scale where resolution limits what's visible
- Obscure prototype roads with less training data than major Class I railroads
- Freelanced or kitbashed models that don't match a real prototype
For these cases, enter the data manually. The AI skips to a manual entry form automatically if confidence is very low.
AI + Human = Best Results
Think of the AI as a first pass — it does the tedious data entry so you can focus on reviewing accuracy. The more you correct the AI's mistakes, the better the platform gets for everyone (see Improving AI Accuracy).
Privacy
Your photos are processed on RailScanPro's secure servers. They are not shared with third parties or used to train AI models sold to other companies. Photos you upload belong to you — see Export Your Data to download your full photo archive.
Next Steps
- Photo Tips — get better results from your photos
- Improving AI Accuracy — contribute corrections to improve recognition
- AI Not Recognizing Items — troubleshoot poor results