You are running for a train, carrying luggage, and the departure board is full of unfamiliar names and times. Translating every word is useful, but your real question is relational: Which services connect, where do I change, and which platform should I check?
We tried that workflow in GPT Realtime Assistant, our voice and photo assistant for iPhone, iPad, and Mac. We attached a prepared sample timetable image to a real app session and asked for a route. This is a practical demonstration with one illustrative timetable, not a benchmark against other products or a test of live station information.
Why translating the words is only the first step
A camera translator can help you understand labels such as departure, arrival, and platform. A traveler still has to match rows across the connections, compare arrival and departure times, and check whether each transfer is possible. Typing all those rows into a chat is another option, but it can be awkward while juggling a phone and a suitcase.
Attaching an image and asking a direct question is a useful alternative: “Which route reaches the destination before 10:30, where do I change, and which platforms are shown?” It does not eliminate the need to check the answer.
The train timetable demo
What the video actually shows: the station scene and presenter are AI-generated, and the timetable is illustrative. For the test, we attached the prepared timetable image as a file to a real GPT Realtime Assistant session; the app response in the video is edited for time. The video is not a recording of an actual camera capture at a live station.
There is also a 13-second edit of the train demo.
What worked in this sample
The illustrative board lists three legs from Midoricho to Kaze Onsen. In the demonstrated response, the assistant identified a route using A12, B24, and K16, with arrival at 10:24. The sample board shows:
| Leg | Departure | Arrival | Departure platform on the sample board |
|---|---|---|---|
| A12: Midoricho → Aoba | 09:00 | 09:23 | 3 |
| B24: Aoba → Sora | 09:30 | 09:52 | 5 |
| K16: Sora → Kaze Onsen | 10:01 | 10:24 | 2 |
Reading the rows together gives a seven-minute connection at Aoba and a nine-minute connection at Sora. The point of the demo is that the assistant can use an image and a question as context for an answer, instead of merely translating labels. We have not run a controlled speed or accuracy comparison with camera translators or other chat tools, so this example does not establish that it is more accurate or faster than them.
What still needs a human check
It is not flawless. Blurred numbers, cropped rows, small footnotes, dates, service restrictions, and ambiguous station names can change the answer. The model can also misunderstand a connection or state an incorrect platform. A photograph captures a moment; it cannot by itself reveal a later cancellation or platform change.
Before acting on a suggested route:
- Check the service date and any symbols or weekend-only footnotes.
- Compare the departure time and platform with the live station board or an official rail source.
- Check that there is enough time to change platforms, especially with luggage or accessibility needs.
- If the image is unclear or the answer is uncertain, ask station staff.
The fictional route and stations above are for demonstration only. Do not use them to plan an actual journey.
A second demo: conversation context
The longer demo shows a different set of situations: a Japanese family conversation, a lecture point turned into a task, and interview preparation.
Its scenes and voices are AI-generated, and some app text was adapted for the demonstration. The family translation Short and technical interview Short are also available.
GPT Realtime Assistant's live voice path uses the cloud; sessions on iPhone and iPad need an internet connection. On supported Macs the app also offers downloadable local Whisper Large v3 Turbo and Qwen 3 8B variants. Those local models are a separate workflow; this train demo is not evidence that its photo reasoning ran offline.
Try the workflow
- GPT Realtime Assistant product home
- How to ask AI about a photo or screenshot
- Travel translation guide
- Product comparisons
- GPT Realtime Assistant on the App Store
The app is free to download; its App Store listing says a subscription is needed for use. Check the listing for current details. When using it to record conversations, ask participants first and follow the rules that apply where you are.
