PDF to Markdown with tables and images: what converts, what breaks

People search for PDF to Markdown with images when a report’s numbers sit in grids and the claims sit in charts. Conversion can pull headings, lists, and some table text into a .md file. It rarely ships a perfect visual twin. Pipe tables break on merged cells and multi-page layouts. Figures often leave captions or empty spots instead of reusable image files. Charts do not become clean data. This article maps what converts, what breaks, and how to check the result before you paste it into notes, Obsidian, or a RAG index. For a text PDF, start at PDF to Markdown. For a photo of a page, use Image to Markdown.

Why tables and figures break when you convert

A PDF stores drawing instructions. Text may live in a readable order. Tables may be lines and positioned cells with no real grid object. Figures may be embedded bitmaps, vector drawings, or a full-page scan. A converter guesses reading order and structure, then writes Markdown. That guess is lossy.

You feel the loss on the parts that were not plain text to begin with. A three-column comparison table can land as one long line. A chart that shows “revenue up 12%” may vanish while the surrounding paragraph survives. A logo in the header can inject noise or nothing at all. The failure is layout, not your editor.

Teams often expect “PDF to Markdown with images” to mean a folder of PNGs plus a .md that links them. Many online paths return Markdown text only. Toolsy’s convert API returns text you can copy or download as .md. Plan for that output shape. If you need the original figure files, keep the PDF or export assets from the design tool that created them.

Markdown is still the right intermediate for editing and LLM work. You get headings you can chunk and tables you can repair by hand. You do not get a print-ready facsimile. Treat the download as a draft of the claims, not a gallery of the artwork.

What usually converts cleanly

Digital PDFs with a real text layer convert best. Clear #-style section titles in the source, short paragraphs, and bullet lists tend to survive. Simple tables with one header row and uniform columns often become Markdown pipes you can fix in a few minutes.

Body copy that was typed in Word or InDesign and exported to PDF keeps character order better than a phone photo of a printout. Page numbers and running headers still leak into the text. You delete those in the .md. That cleanup is cheaper than rebuilding the whole document.

Link text and basic emphasis sometimes appear. Complex styling (multi-level drop caps, text boxes floating over photos) usually collapses into linear paragraphs. That is fine for RAG and notes. It is a problem if you needed the exact brochure layout.

Run a sample of three pages before you batch a binder. If those pages keep headings and one clean table, scale up. If they scramble columns, fix the source export or switch strategy for that file class. For why models prefer this intermediate at all, see why LLMs prefer Markdown over PDF.

Tables: what becomes Markdown pipes

Converters try to rebuild | columns from cell positions. Success depends on how the PDF encoded the grid. A real table object or consistent cell text is easier than decorative lines drawn around floating words.

Expect to open the .md and count columns against the PDF. One wrong pipe shifts every value to the right. Finance and compliance tables need that check every time. Marketing comparison grids need it when you will quote a number later.

If the “table” still exists as a spreadsheet, skip the PDF. Export .xlsx and use Excel to Markdown. You start from rows and columns instead of a printed picture of them.

Simple grids and header rows

A small grid with a single header row and no merged cells is the best case. Conversion often yields a Markdown table you can paste into GitHub, Obsidian, or a docs site. Spot-check one data row and the header spelling. Fix spacing so each row has the same number of cells.

Lists that look like tables in the PDF (tab-aligned columns without rules) may become plain paragraphs or a broken table. Re-type those as real pipes if the values matter. For RAG, a short corrected table beats a long garbled line that embeds as noise.

Wide, merged, and multi-page tables

Wide tables wrap or split. Merged header cells collapse into duplicated or empty slots. Tables that span pages often become two broken fragments with a footer stuck in the middle. You will spend more time repairing those than converting them.

Keep a spreadsheet export as source of truth when the grid is the product (pricing matrix, clinical schedule, multi-year budget). Use PDF conversion for the surrounding narrative. Point readers or your index at both artifacts when a question needs exact cells.

People who search “extract tables from pdf” often want Excel or CSV, not Markdown. That is a different job. Markdown pipes are for docs and prompts. Spreadsheets are for calculation. Pick the output that matches the next step.

Images and figures: what you actually get

“PDF to Markdown with images” mixes two wishes: keep the pictures, and keep the meaning of the pictures. Most text-first converters help more with meaning than with asset files. Captions and nearby paragraphs may survive. The bitmap itself may not land as a portable ![](...) you can open offline.

Toolsy’s PDF converter returns Markdown text. It does not promise a ZIP of extracted figures. If a run includes image placeholders or alt-like scraps, treat them as hints, not a media library. Keep the original PDF beside the .md when you still need the visual.

Microsoft’s MarkItDown library family can do more with OCR plugins and LLM vision when you configure them locally. A hosted one-click convert is a different product surface. Do not assume every MarkItDown feature from GitHub is on in the browser path. For library vs online framing, see MarkItDown explained.

Captions versus image files

Captions under figures are text. They convert like any other paragraph when the PDF has a text layer. That caption is often enough for search and RAG: “Figure 3 shows defect rate falling after week 12.” Write a one-line claim under the heading if the caption is thin.

Image files are separate. Designers who need the PNG for a wiki should export from the source deck or use a PDF tool that explicitly unpacks embeds. Expect an extra step. Do not block your Markdown pipeline waiting for perfect asset export if the goal is editable prose.

Charts, diagrams, and equations

A chart is a picture of data. Conversion does not reliably rebuild the underlying series as a table. Axes labels may OCR poorly. Legends may shuffle. If the chart carries the claim, type the claim in plain sentences after conversion, or pull numbers from the spreadsheet that built the chart.

Flow diagrams and architecture sketches lose arrows and grouping. Equations and handwritten formulas need specialized tools or a human rewrite. Obsidian users who want equations and images together often combine a converter with a separate image extract; that workflow is heavier than a single upload. A later guide covers importing PDFs into Obsidian as Markdown.

How to prepare the PDF before conversion

Start with a text PDF. If you still have the Word, Google Doc, or deck, export fresh instead of scanning a printout. Flatten passwords so the converter can read the file. Drop cover art pages that hold no prose.

Crop or remove pages you will not retrieve. A 200-page binder with two useful tables wastes quota and QA time. Split by chapter when headings already match how you retrieve.

Straighten scans. Raise contrast. Prefer 300 DPI over blurry phone photos. Even with OCR, skewed pages scramble columns. If the whole file is image-only, plan an OCR path and a longer proofread. Name the file with a stable ID so your .md and source stay paired in storage.

Skip public upload for secrets, regulated health data, or files your policy forbids on third-party servers. Use an approved local pipeline for those. For upload norms on Toolsy, the privacy sibling posts cover retention and processing in more depth.

Convert, then check tables and figures by eye

Upload the PDF on PDF to Markdown. Wait for the Markdown text. Download or copy the .md into your editor. Free accounts get a daily file cap and a size limit around 5 MB; Pro raises those caps.

Skim headings first. Confirm section titles match the PDF outline. Delete repeated headers and page numbers. Then open every table you care about. Compare one header and one numeric row to the source. Fix pipes until column counts match.

For figures, search the .md for “figure,” “table,” and known chart titles. If the claim is missing, add a sentence under the right heading. If you only have a photo of a page, run that photo through Image to Markdown and merge the text by hand.

Save the cleaned .md next to the PDF. Re-convert when the source changes; do not maintain two conflicting edits without a rule for which file wins. For corpus-scale prep, follow how to prepare documents for RAG with Markdown.

When image OCR or Excel export beats PDF conversion

Use Image to Markdown when the source is a JPG, PNG, or WebP of a page, whiteboard, or slide. That path is OCR into Markdown. It works best on sharp printed text. Messy handwriting fails more. It does not replace a text-layer PDF when you already have one.

Use Excel to Markdown when the grid still lives in a sheet. You avoid reconstructing cells from ink. Wide pricing tables belong here.

Use a heavier local stack (Marker-class tools, MarkItDown with OCR plugins, or a dedicated PDF table library) when you batch complex layouts and can run Python on your own machine. Product comparisons in that space belong in Marker vs MarkItDown vs online converters. Online conversion stays the right default for a clean report you need in minutes.

Limits, privacy, and when to stop

Online conversion is for files you are allowed to upload. Toolsy processes the file to produce Markdown and does not keep it as a library copy after the response, per the tool FAQ. That still means the bytes leave your device for the job. Local CLI wins when policy says so.

Do not expect pixel-perfect figures, perfect merged cells, or chart-to-CSV magic. Do not treat OCR digits as audited without a human pass. Stop and change approach when the PDF is mostly photos, when DRM blocks reading, or when a single wrong cell would cause real harm.

Quota and size limits matter for batches. Split large binders. Prefer source exports under the size cap. If conversion returns empty text, the file may be image-only or corrupt; try OCR on a page image or fix the export.

Honest limits keep the Markdown tree useful. Overclaiming fidelity is how bad tables enter production RAG.

Frequently asked questions

Does PDF to Markdown with images keep the pictures?

Usually you get text, captions, and structure, not a full set of image files ready for a gallery. Many converters, including Toolsy’s PDF path, return Markdown text rather than a ZIP of PNGs. Keep the original PDF when you still need the visuals. Export figures from the design source if you need separate assets.

Can I convert PDF to Markdown with images and tables in one pass?

You can convert once and review both. Tables may become Markdown pipes. Images are the weaker half. Plan a QA pass for every grid you will quote and a caption or rewrite for every chart that carries a claim. One pass is a draft, not a finished knowledge base entry.

How do I get PDF to Markdown with tables that stay accurate?

Prefer a text PDF or an Excel export. After conversion, compare header cells and one full data row to the source. Rebuild merged or multi-page tables by hand or from the spreadsheet. Accuracy comes from that check, not from hoping the converter inferred every rule line.

What breaks first on complex PDF tables?

Merged cells, multi-page grids, and decorative column layouts break first. Footers land inside rows. Wide tables wrap into uneven pipe counts. Those cases need spreadsheet export or manual rebuild more than another converter click.

Is extract tables from PDF the same as Markdown conversion?

Those queries often mean Excel or CSV for analysis. Markdown tables serve documents, README files, and model context. If you need formulas and filters, export to a sheet. If you need readable docs, Markdown pipes are enough after a cleanup.

Should I use image to Markdown for a scanned PDF?

If the PDF is a scan or you only have a photo of a page, OCR is required. A single clear page image can go through Image to Markdown. A multi-page scan may need OCR-capable PDF tooling or page-by-page images. Expect proofreading on names and numbers either way.

Will charts become data tables in the Markdown?

Rarely in a reliable way. You may get axis fragments or nothing useful. Type the claim in prose or copy values from the sheet that created the chart. Treat chart bitmaps as illustrations until a human verifies the numbers.

Does MarkItDown preserve figures better than an online converter?

The open-source MarkItDown stack can add OCR and LLM vision when you install plugins and supply a model client. That is a local engineering setup. A browser converter optimizes for upload-and-download text. Compare goals before you assume feature parity. See the MarkItDown guide for library vs online.

Is it safe to upload a PDF that contains figures and internal tables?

Only if your policy allows that file on a third-party service. Prefer tools that process for the response and do not retain the upload as a stored archive. Keep confidential material on an approved local path. When in doubt, redact or withhold the file.

What is the fastest path for a clean text PDF with one simple table?

Upload it to PDF to Markdown, download the .md, fix the table pipes, and delete header junk. That path beats photographing pages or chaining print-to-PDF from an app that can still export Word or Excel.

For RAG-scale cleanup, continue with prepare documents for RAG with Markdown. For model-side rationale, see why LLMs prefer Markdown over PDF. For library vs browser tooling, see what MarkItDown is.