GST invoice OCR for scanned papers and mobile photos

By Dynamite Docs. Published 2026-08-14. Updated 2026-09-11.

Turn GST invoices into clean, reviewed data for Excel and Tally.

GST invoice OCR is for files that have pixels but no reliable text layer: paper scans, photocopies and mobile photos. Dynamite Docs uses a vision-capable model to propose the printed GSTIN, [invoice](/accounting/india/tally-invoice-extraction) identity, HSN or SAC, line items, taxable value and tax components as structured fields. The source image stays available during review so a person can inspect faint digits, cropped edges and shifted table rows. Export only after the GSTIN roles, arithmetic and tax split agree with the visible invoice.

Scope and tax review

Content owner: Dynamite Docs content team. Official-source review last updated: 2026-09-11.

Dynamite Docs extracts fields from source documents. It does not determine whether an invoice complies with GST law, prepare a return, or make a document ready to file. Review every extracted value against the original and use the current GST Portal guidance or a qualified tax professional before filing.

We publish the controls we actually operate rather than claiming certifications we do not yet hold.

Official GST sources

CGST Rules: tax invoice particulars (Rule 46), Central Board of Indirect Taxes and Customs. Mandatory 16 invoice particulars and tax invoice format requirements.
Form GSTR-2B viewing and reconciliation FAQs, GST Portal. Auto-drafted ITC statement, downloads, cutoff dates, and book reconciliation.
Form GSTR-1 preparation and Returns Offline Tool FAQs, GST Portal. GSTR-1 invoice data, preparation modes, validation, and error correction.
Steps for e-invoicing, GST Portal. IRP reporting, IRN, B2B QR code schema, and GSTR-1 auto-population.

Try a GST invoice or compare its storage, processing, and export limits.

The actual Dynamite Docs library import menu, with upload and Google Drive options.

The actual AI processing dialog showing the searchable model list and provider filters.

A purchase order beside its extracted text in the Text Editor.

A purchase order beside its structured details in the Table Editor.

A purchase order beside its extracted line items in the Table Editor.

A purchase order beside the Table Editor Totals tab, showing tax and the final total.

Purchase order text beside extracted document fields and confidence indicators.

The actual Export to Google Drive dialog with format, scope, file name and folder options.

GST invoice OCR for scanned papers and mobile photos: sample source document and illustrative field names and values — **01 / Sample field guide** — Illustrated field guide. Example values may differ from the downloadable sample.

Dynamite Docs product screenshot with a vendor invoice beside extracted fields — **02 / Product review workspace** — Invoice example showing the shared review controls. Fields vary by document type.

Paper bills and photo captures slow down bookkeeping

Scanned papers and photos contain image pixels rather than selectable text, requiring someone to read and retype every single number.
Photos sent over messaging apps often have glare, shadows, angled perspectives, or creased paper.
Important fields like the 15-character GSTIN, HSN/SAC code, and tax amounts are often printed in small fonts or faint thermal ink.
Basic OCR software produces unorganized text blocks rather than the clean spreadsheet table your accounting workflow requires.

Document types this workflow handles

Scanned GST invoice PDFs: paper invoices scanned to PDF without selectable text
Photocopies & faxes: faded and low-contrast paper documents
WhatsApp & mobile photos: photos taken on phones with varying angles and lighting
Thermal paper purchase slips: fuel, retail, and local supplier receipts

What gets extracted from a scanned GST invoice

The same comprehensive field set as a digital PDF, with confidence scores highlighting any unclear reads:

GSTIN (supplier): 07AAECS4458H1ZF
Invoice number: GT-8821
Invoice date: 2026-08-05
Vendor / customer: Vardhan Packaging
Place of supply: Delhi (07)
HSN / SAC code: 4819
Item & quantity: Corrugated boxes x 800
Taxable value: ₹36,000.00
CGST (rate · amount): 9% · ₹3,240.00
Invoice total (incl. tax): ₹42,480.00

Tax invoice, Mehta Steel Supplies, INV/2026-27/0188. Fictional sample document. — Downloadable sample: INV/2026-27/0188. Demonstration data for extraction practice.

A realistic example

Sample data. Not a real invoice or filing record.

A batch of photographed and scanned invoices extracts into clean rows like these, with confidence flags on unclear numbers:

Invoice #	Date	Vendor	GSTIN	HSN/SAC	Item	Taxable value	CGST	SGST	IGST	Total
GT-8821	2026-08-05	Vardhan Packaging	07AAECS4458H1ZF	4819	Corrugated boxes	₹36,000	₹3,240	₹3,240	Not applicable	₹42,480
KT-5512	2026-08-07	Ganesh Industries	29AABCC3456C1ZD	8467	Rotary hammer	₹1,86,000	Not applicable	Not applicable	₹33,480	₹2,19,480
TH-0916	2026-08-08	Prabhat Hardware	27AAEHP5621L1ZV	7323	Steel pipes	₹28,000	₹2,520	₹2,520	Not applicable	₹33,040

What to validate before export

Validation issue

Low-quality scans need character-level review of GSTINs, dates, and amounts.

Limitation

OCR output is not a compliance validation of the invoice.

Reviewed by Dynamite Docs content team. Last verified 2026-09-11.

Optical character recognition designed for Indian invoices

OCR quality depends on what the image actually contains. Glare can erase a decimal point. Compression can turn one character into another, while a page photographed at an angle can pull an amount into the next column. A high-confidence read is useful, but it is not evidence that the tax value is correct. Check identifiers and money against the image.

Start with document identity. Confirm supplier and recipient GSTINs, invoice number, date and document type. A GSTIN format check can catch the wrong length or character pattern, but it cannot prove that the number belongs to the party shown. Keep the role label and source available to the reviewer.

Move to line items only after the header is settled. Check HSN or SAC, description, quantity, unit, rate, discount, taxable value and tax rate. Long descriptions may wrap under the next line, and faint grid lines may cause column shifts. Compare the extracted line count with the invoice before trusting the total.

Recalculate the summary using the printed rows where possible. Freight, insurance, rounding, discounts and cess may sit outside the main item table. Record those components rather than changing a line to force the arithmetic to agree. The reviewer should also confirm whether the document shows CGST with SGST or UTGST, or IGST.

For a better source image, photograph the whole page on a flat surface, keep the camera parallel to the paper and avoid shadows across the totals. Do not crop out the invoice heading, GSTIN labels or final amount. If a scan is unreadable to a person at normal zoom, treat OCR as a draft and ask for a clearer copy.

Test several difficult documents before setting up a batch: a faded printout, a multi-page invoice and a mobile photo. Agree on which confidence levels trigger extra review, but never skip GSTIN, tax and total checks solely because a field scored highly. Dynamite Docs prepares visible data; your finance or tax professional remains responsible for compliance and filing.

Flexible vision models with transparent pricing

Document quality varies between clean digital files and faded paper scans. Use a hosted model or connect a supported provider key. Pro and Ultra BYOK runs do not spend Dynamite Docs PE, though provider and file-size limits still apply.

A digital PDF may provide machine-readable text. Scans and photos need a vision-capable model, but both formats need structured review for line boundaries, GSTIN roles and tax values.

All extracted data and confirmed supplier corrections stay in your private workspace, ready to export whenever you need.

Questions, answered

Can it read faded thermal receipts or crumpled bills?

Yes. Optical character recognition reads low-contrast printouts, faded thermal slips, and paper copies. Any unclear digits are flagged with a low confidence score so you can verify them.

How are clean digital PDFs handled?

A digital PDF can supply machine-readable text, which avoids image OCR. The document still goes through structured extraction and review, subject to the selected processing option and plan limits.

Which vision AI providers can I use?

You can use our built-in models or connect your own API keys from Gemini, Mistral, OpenAI, Anthropic, Groq, Cloudflare, or OpenRouter. On Hobby, Pro, and Ultra plans, you can also run model inference locally using Ollama.

Can it handle photos taken at an angle with mobile phones?

You can submit a mobile photo, but angle, shadows, glare and cropping affect the result. Retake poor images when possible and compare every highlighted identifier and amount with the source.