Drop a chart. Get a read.
Two reads from one image, and the screen always says which you are holding. Measured: the ticker comes off your picture and the levels come off fetched candles. Beta: a model reads the picture itself and tells you what it thinks — inferred, never graded.
The beta is the exception to how everything else here is built, and it ships labelled rather than quietly: no beta read is graded, journalled, or counted in any published accuracy figure. Those numbers describe measured reads, and they stay that way.
Where the two reads stand
Drag, paste with ⌘V, or use your camera roll. This page previews your file and explains both reads; running one needs an account, because a read spends a daily analysis.
Drop a chart screenshot
Drag it here, paste with ⌘V, or tap to pick from your camera roll — and see exactly what ORIN takes from an image, and what it refuses to.
Choose an imageProcessed in your browser · never uploaded from this page
This dropzone previews your file locally and sends nothing — it cannot run a read. Inside the desk an upload is downscaled and sent to be read, either for its ticker alone or, in the beta, for the model's own view of the picture. The sample plan alongside is a worked example on a different symbol, labelled as one.
One image. The measured read, and the beta's.
The left column is what the Upload beta gives you: a model's reading of the picture. The right is the measured read on the same instrument. Both ship, both are labelled, and only one of them has data behind it — which is the whole reason they are shown side by side rather than blended into one answer.
Upload beta
UngroundedReads the picture and states what it sees
"Sell-side liquidity swept near 124.10, reclaimed within two candles."
Plausible, and read off an axis in an image. Nothing checked it against a candle.
"Invalidation around 124.90, beneath the block that has to hold."
A level someone drew on the chart looks the same to a model as one the market made.
Works on any instrument, including ones no vendor sells us data for.
The reason the beta exists. A futures or exotic chart has no measured read at all.
Never graded, never journalled, never in a published accuracy figure.
Not a limitation of the beta — including it would make those figures untrue.
Measured read
MeasuredTakes the ticker, then grades real candles
Sell-side liquidity swept at 124.10 and reclaimed within two candles.
Computed from the vendor’s bars. The evidence id behind it is citable on the read.
Invalidation at 124.90 — beneath the order block that has to hold.
Derived from measured structure, not from a round number or a drawn line.
Only for instruments ORIN can fetch. It says so, and names the proxy, when it cannot.
A refusal is an answer. Guessing the instrument would be the easy path and the wrong one.
Graded, and its confidence states what it rests on — a resolved sample, a replay, or nothing yet.
Most bands have no live sample behind them yet, and the read says so in those words rather than printing a bare number.
A general model gives you the left column and calls it the right one. The difference here is not that ORIN refuses to read a picture — it does — but that it never lets you mistake one for the other.
What happens to your image
Three steps, and each has an honest edge case that is surfaced on the read rather than hidden here. The edge cases are the useful part of a feature that reads pictures.
Downscale
Your file is resized to a 1024px longest edge and re-encoded in your browser, then uploaded. It is read and not kept — but it does leave your device, and this page will not tell you otherwise.
Edge case: a photograph of a screen rather than a screenshot loses the axis labels first. That is the text the whole read depends on, in both modes.
Choose the read
Measured takes only the ticker and timeframe off the image and grades that market’s own candles. Beta hands the whole picture to a vision model and returns its reading of it. One control, and the screen never stops saying which you picked.
Edge case: the measured read declines instruments no vendor sells us — futures, most exotics — and names the proxy instead. The beta will read any chart, including those, with nothing behind it.
Read it beside the other one
When the ticker resolves, the beta read carries a button that grades the same instrument on fetched candles. Two answers, one screen, and the disagreement between them is worth more than either alone.
Edge case: an image is a frozen moment and carries no timestamp we can trust. A screenshot from last week reads exactly like one from a minute ago — the measured read is the one that knows what time it is.
What the beta gets wrong
An image is a frozen moment read by a model, and these are the specific ways that fails. They are published because a feature that reads pictures is only safe to use if you know exactly what it cannot see.
The levels may not exist
A price axis in a screenshot is a few dozen pixels of text. It can be misread, read off a secondary scale, or inferred where it was cropped. A beta level is a number to check against your own chart, never one to size a position from.
Drawings are pixels too
A line somebody drew looks the same to a model as a level the market made. The beta may build a whole read around a trendline from a Telegram chat, and it has no way to know it did.
The image has no clock
There is no live price in a picture and no timestamp worth trusting. A two-hour-old screenshot produces a two-hour-old read that reads as current, and price does not care that you only just asked.
For a live symbol, use the measured read
If the ticker is one we cover, the measured read pulls current data and gives you current levels — strictly better than a model’s reading of a picture of the same chart. It is one click from the beta read, on the same screen.
The honest recommendation is unchanged by the beta existing: use the AI chart scanner whenever the instrument is one we can measure, and reach for the beta when it is not — an instrument with no data behind it, or a chart you simply want a second opinion on. Pointing you at the other tool costs us nothing and saves you a worse answer.
Questions traders actually ask
It depends entirely on whether the model measures the image or describes it, and ORIN now offers both — labelled. The measured read takes only the ticker and timeframe off your image and grades that market’s real candles, so every price traces to data. The Upload beta lets a vision model read the picture itself and state its own levels; it is often useful and it is a guess about pixels, which is exactly what the screen it appears on says.
You can, and it will answer confidently. The difference is not that ORIN refuses to read a picture — the Upload beta does read it — but that ORIN tells you which kind of answer you are holding. A general model gives you one output and no way to know whether a level was found or invented. Here the inferred read is marked as inferred, the measured read is available on the same screen for any instrument we cover, and you can put them beside each other.
Yes, in two modes on one screen. Measured read: the model reads only the instrument and timeframe, and the ordinary pipeline grades that market’s candles — no level comes from your pixels. Upload beta: the model reads the chart itself and returns a direction, levels, an entry, an invalidation and its reasoning, all inferred from the image. Beta output is never graded, never added to your journal, and never counted in any accuracy figure we publish.
They are the model’s reading of a price axis in an image, which is a few dozen pixels of text. They can be off, read from a secondary scale, or inferred where the axis was cropped — and a level someone drew on the chart looks the same to a model as one the market made. Treat them as a starting point to check, not as measurements. If the instrument is one ORIN covers, the measured read is one click away on the same screen and it is the one with data behind it.
In the measured read they are ignored entirely — only the ticker and timeframe are taken from the image, so a line you drew cannot influence a grade. In the Upload beta the model sees them exactly as it sees the candles and may well build a read around one, which is a real limitation of reading a picture and is stated on the screen rather than buried here.
No. The image is downscaled in your browser, sent to be read, and not kept after the read returns — in either mode. It does leave your device, which is why this page does not claim otherwise. The dropzone on this page is a different thing: it previews your file locally and cannot run a read at all, because a read spends a daily analysis and needs an account.
The first read is free with no account, on a symbol. Uploads need an account because each read — measured or beta — spends one of your daily analyses, including when the image turns out to be unreadable, since the model was run either way.