Step 1
Camera access
A short instruction: take the document out of its cover, find some light. Then the camera request. On a computer without a suitable camera, the user continues on a phone via a QR code.
Document recognition from a video stream
Recognition of documents from a phone or computer camera video: the service finds the passport in the frame, prompts “closer” or “remove the glare”, takes the shot itself once the document is in the frame and sharp, and recognises the fields at once. Photos of a screen and printouts are rejected at capture time.
Live capture Russian passport · main spread
- Frame in frame The whole document is in the frame
- Glare no No glare on the fields
- Sharpness sharp The frame is sharp — the shot was taken automatically
- Result recognised Fields recognised: Lesnyak Agafya Mironovna
- Prompts before the shot, not after a rejection
- Up to 6 free attempts per step
- 16,000+ document types from 195+ countries and territories
Data is stored in Russia Russian Software Registry No. 19972 Compliance with 152-FZ
Document recognition from a video stream: what changes when the camera is on
Fewer retakes — fewer abandoned checks
With a regular photo upload, the error is visible only after submission. In the camera stream, the check document recognition starts earlier: the service looks at the frame while the person holds the phone and helps to capture the document at the first attempt.
Every prompt in the stream saves an attempt: only the step that had a problem is retaken.
- 01
A frame instead of guesswork
The screen shows a frame for the document. The user sees where to put the passport and from what distance to shoot, instead of picking an angle at random.
- 02
Advice before the shot
“Closer”, “further”, “more light”, “move your finger off the field” — the prompt appears while the camera is on, not after a rejection on the server.
- 03
The service picks the frame
When the whole document is in the frame, still and sharp, the shot is taken automatically. Perspective is straightened, glare and noise are removed before recognition.
- 04
Live capture only
The document is captured by the camera, not picked from the gallery. Photos of a screen, photocopies and printouts are rejected with a clear reason.
- up to 6 free attempts per step
- 16 000+ document types
- 195+ countries and territories
From the camera preview to JSON fields: six capture steps
This is what one “Document” step looks like in the Web SDK or mobile SDK — from camera permission to recognised fields on your server. What the process looks like on a phone — in the Web SDK documentation.
Step 2
Document in the frame
A camera preview with a frame for the required document. For weaker devices the frame can be made vertical or horizontal. The phone is held vertically.
Step 3
Frame check in the stream
The service checks that the document is entirely in the frame, large enough, lit, free of glare on the fields and in focus. If something is wrong, it prompts at once.
Step 4
Automatic shot
When all conditions are met and the document is still, the frame is taken automatically. There is no need to press a button and guess the moment.
Step 5
Recognition and anti-fraud
The frame goes to the server: document type and country, alignment, fields, MRZ checksums, checks for a screen, a printout and signs of editing.
Step 6
Fields to your server
The user sees “Document accepted” and moves to the next step. The recognised fields and the step status arrive by webhook or on a session status request.
What the service checks in the frame before submission
Six conditions of a valid frame. While at least one is not met, no shot is taken and the user sees what to fix.
-
Whole document
All edges of the form are in the frame, nothing is cropped. Fields cannot be read and security features cannot be checked on a cropped document.
-
Large enough
The document takes up most of the frame rather than lying as a small spot on a table. Otherwise small text is not recognised.
-
Enough light
A frame that is too dark is rejected: the prompt suggests turning on a light or moving to a lit place. It is better not to use the flash — it causes glare on the lamination.
-
Fields uncovered
Neither glare, nor a finger, nor a cover hides the surname, number, dates or the machine-readable zone.
-
Sharpness
The frame is in focus, without blur from hand movement. A blurry shot gives low-confidence fields and MRZ checksum errors.
-
Colour without filters
A colour frame without processing: a black-and-white image looks like a photocopy and is not accepted.
Prompts before the camera: what advice the user gets
- Document is far away
Bring the camera closer to the document. - Document does not fit
Hold the camera further away: the whole document must fit in the frame. - Phone is rotated
Turn the device to portrait mode.
- Too dark
Turn on extra light or move to a better-lit place. - Glare on the fields
Tilt the document or the phone so the glare leaves the fields; do not use the flash.
- Blur and defocus
Wipe the camera and hold the phone still until the shot is taken. - MRZ lines are unclear
Take a new shot: the lines at the bottom of the passport must be fully visible and sharp. - Weak camera
Try again from a more modern device or continue on a phone via a QR code.
- Step not passed
Only this step is retaken, the others stay accepted. - Last attempt
After the sixth failed attempt on a step, the session is closed and a new one is needed.
- Prompt texts can be shown as they are or rewritten in your own words in the scenario editor. Below — what the person sees and when.
- “In the stream” — a prompt before the shot; “After the shot” — a server response with a code. The full list of codes and texts is in the error reference.
- Why a document is of the wrong type or the wrong page — on the page document classification.
Camera stream or a ready image: which channel to choose
Live capture is the best path when the client is next to the camera. For scans that are already in your system, there is the REST API. Request parameters are in the method documentation.
- Enable
Web SDK
camera stream in the browser
When it fits
- registration on a website
- a link in an email or message
- a computer without a camera — continue on a phone via a QR code
What matters
- page over HTTPS
- widget on an allowed domain
- no file upload in the widget
- Enable
SDK for iOS and Android
camera stream in an app
When it fits
- a mobile app of a bank, a rental service, a marketplace
- an employee’s tablet at a point of service
What matters
- camera permission for the app
- portrait orientation
- styling — in the scenario
- More — mobile SDK
- Enable
REST API
ready images without a stream
When it fits
- scans from the back office
- documents from email and CRM
- archive
What matters
- JPEG or PNG, up to 10 images per request
- no prompts at capture time
- protection against substitution is weaker than with camera capture
Screen, printout, gallery shot: anti-fraud at capture time
-
Document on a screen
A passport opened on a monitor, tablet or another phone is rejected: the service sees a screen in the frame and asks to shoot the original.
-
Photocopy and printout
A black-and-white copy and a colour printout instead of the original are rejected. If you need scans and copies, these checks are turned off in the scenario.
-
Shot from the gallery
The widget has no “upload a file” button: the document is captured by the camera here and now. A pre-prepared or edited shot is harder to slip in.
-
Face as a separate step
After the document, the scenario can ask for a selfie with the document and Liveness: a live person in front of the camera and a one-off match with the passport photo within this check.
Some substitutions are easiest to stop while the person is holding the camera. Above — what is caught when a document is captured in the stream.
Pasted-in elements, erasures, traces of a graphics editor and logical inconsistencies are checked on the server — more on the page authenticity check. Substitution of the video stream itself at the face step is covered by deepfake protection.
Where reading a document through the camera replaces scans
Wherever the client has the document in hand and a shot error costs an application or a queue.
-
Banks and MFIs
- An application from a phone: the passport is captured in the stream, fields fill in the form without manual entry.
- Example: Passport → application form
-
Car sharing and rental
- Both sides of the licence are captured by prompts, the reverse side at no extra charge.
- Example: Driving licence: front and back
-
Points of sale and issue
- An employee points a tablet at the client’s passport instead of retyping the data.
- Example: Passport at the counter
-
Hotels
- A guest’s passport is captured at the front desk or before arrival via a link, the data goes straight into the system.
- Example: Passport or ID card
-
Couriers and contractors
- Registration from a phone on the road: prompts help to capture the document in poor light.
- Example: Foreigner’s passport + patent
-
Telecom operators
- SIM card activation and tariff changes: the subscriber’s passport — by camera in the app or in a store.
- Example: Subscriber’s passport
Camera frames and 152‑FZ: what goes to the server and how long it is stored
Only the frame needed
A camera stream is not a video recording of the client
Under Part 5 of Art. 5 of 152‑FZ, data must not be excessive. For the “Document” step, a photo of the document is sent to the server, not a video recording of whatever got into the camera.
Retention period — yours
Set in the scenario
Under Part 7 of Art. 5, data is kept no longer than the purpose requires. The retention period of session data is set in the scenario, and you can delete a session earlier with an API method.
Databases in Russia
The NeuroVision cloud — on servers in Russia
Under Part 5 of Art. 18, the data of Russian citizens is recorded and stored on databases located in Russia during collection. The NeuroVision cloud runs on servers in Russia.
Your perimeter
Frames do not leave your infrastructure
On installation in your environment document frames do not leave your infrastructure — for banks and companies with information security requirements.
The data operator is your company
A guideline, not a legal opinion
This is a guideline, not a legal opinion; the personal data operator remains your company.
152‑FZ
a Russian developer, software in the registry, No. 19972
How much document recognition from a camera costs
Capture in the stream is not charged separately: prompts, frame checks and anti-fraud are included in the package. You pay for a KYC session with all steps, with no subscription fee. Test access — 1 month.
Calculate the cost- 01
From 16 ₽ per session — the “Document” package: quality check, classification, recognition and document anti-fraud
- 02
10 ₽ — an additional document in the same session; reverse sides of licences and ID cards — free; up to 6 free attempts per step
- 03
From 29 ₽ — “Document + liveness”: plus a selfie or Liveness and a face match with the photo in the document
Enabling capture: Web SDK, SDK for iOS and Android, or a link
Capture in the stream works in every method where the NeuroVision camera shoots. You do not need to write your own camera interface.
- Frame and prompts
- Auto-capture
- No camera interface of your own
- Test flow and documentation
Terms
Test access for 1 month, a personal manager, support under contract, pay only for checks.
Glossary of document capture in a video stream
- Video stream
- A real-time image from the camera. The service analyses it while the camera is on, but sends one selected frame for recognition.
- video stream
- OCR in the stream
- Document recognition where capture and quality checks happen in the camera stream rather than on a previously uploaded file.
- real-time OCR
- Document capture
- Finding the document in the frame, checking the conditions and taking a shot suitable for recognition.
- document capture
- Auto-capture
- A shot without pressing a button: the frame is taken automatically when the document is in the frame, still and sharp.
- auto-capture
- Frame
- An outline on the screen into which the user places the document. It sets the distance and angle.
- capture frame
- Glare
- A bright spot from a lamp or flash on the lamination of a document. It covers fields and interferes with recognition.
- glare
- Perspective alignment
- Bringing a document shot at an angle to a rectangular shape before the fields are recognised.
- perspective correction
- Presentation attack
- An attempt to show the camera something other than the original: a screen, a printout, a copy.
- presentation attack
We will show capture in the stream on your documents
Tell us which documents your clients capture and from which devices — we will show the frame, prompts and auto-capture on a demo scenario and give you test access for 1 month.
Can't put your request into words? Book a call and our team will contact you at a convenient time
Book a call
The company is included in the register of IT companies and the register of personal data operators of the Russian Ministry of Digital Development.
Address: Moscow, Presnenskaya Embankment 12 — Moscow City business centre, Federation Tower. Nearest metro stations: Delovoy Tsentr and Mezhdunarodnaya.
Read also
How to capture documents so that they are accepted at the first attempt — in our blog
434 21
KYC Document Verification
Why users retake documents: how to reduce the number of repeat attempts during a KYC check
439 10
AI OCR, IDP and Document Recognition
How to improve the accuracy of AI OCR and MRZ verification in KYC: metrics, errors and working solutions
406 27
KYC, AML and Digital Identification
The KYC Funnel: How to Increase Verification Completion and the Approval Rate Without Slowing Down the Decision
396 23
KYC Document Verification
AI-OCR in KYC: how automatic document recognition reduces manual review
Frequently asked questions about document recognition from video
Answers about capture in the stream, prompts, anti-fraud and cost
It is capturing a document in the camera stream: the service looks for the document in the frame, prompts how to shoot it, takes a valid shot itself and recognises the fields. Quality errors are fixed before submission, not after a rejection.
The user sees at once what is wrong: “closer”, “remove the glare”, “more light”. The document is captured here and now, so a pre-prepared or edited shot is harder to slip in.
No. The service works with the camera stream in real time and sends the selected frame for recognition. Ready JPEG and PNG images can be passed through the REST API.
No. When the whole document is in the frame, still and sharp, the frame is taken automatically.
Yes, if the computer camera is suitable and capture from a computer is allowed in the scenario. Otherwise the widget shows a QR code and the user continues on a phone.
The service notices that glare covers the fields and asks to tilt the document or the phone. It is better not to use the flash: on lamination it causes glare.
No. A photo of a document from a monitor, tablet or phone screen is rejected, and the user is asked to shoot the original.
Up to six free attempts per step. Only the step that had a problem is retaken; after the last failed attempt the session is closed.
Any documents from the list: 16,000+ types from 195+ countries and territories — passports, driving licences, ID cards, residence permits. Permitted documents are set in the scenario.
There is no separate charge for capture in the stream: the “Document” package is from 16 ₽ per session, an additional document is 10 ₽, the first month of testing is free.
