<!-- Generated from pdf-to-slides.md. Do not edit by hand. -->

> Markdown twin of https://conduitllm.com/pdf-to-slides.html
> Turn a PDF report into an editable slide deck with AI: attach the PDF, approve a storyline, build the slides, edit the words and export to HTML, PowerPoint or PDF.

---
Tutorial

# How to turn a PDF into a slide deck with AI

Updated 31 October 2026

The report is written. Now it needs to be a presentation, by Thursday. In [Conduit](https://conduitllm.com/), a free, open-source desktop AI app, you attach the PDF, agree on what each slide should say, and get a deck you can edit, present and export, built from the report's own numbers.

## Introduction

Asking a chatbot to "make slides from this PDF" usually gets you an outline, or a file you can't change without starting over. Conduit's **Slides** work differently. The model first proposes a storyline, one line per slide, which you edit before anything is built. It then builds the slides one at a time with slide tools, so a later change rewrites one slide instead of the whole deck, and each chat turn that changes the deck is saved as a version you can go back to.

In this tutorial, you will turn the five-page annual report of Riverton Bike Share, a city bike-share scheme, into a deck for a city council committee. You will attach the PDF to a chat, approve a seven-slide storyline, build the deck, edit its words in the Script view, ask for an extra chart slide from the report's finance table, and export the result.

The deck in this tutorial was made with GLM 5.3 Flash, a fast, low-cost model, through OpenRouter. The storyline, the eight slides and one change cost $0.0035 in total. You can follow along with any PDF: a report, a research paper, a project plan or a long proposal.

## Key takeaways

- Attach the PDF to a chat and ask for a deck. The chat turns into a deck and opens in **Slides**.
- You approve a storyline, one line per slide, before any slide is built.
- The deck is built slide by slide from the PDF's text, with charts and stat rows where the numbers call for them, and a layout check after each change.
- The **Script** view shows every word on every slide as one editable document. Lines you edit yourself are kept when the AI changes the deck.
- **Export** saves an HTML file that presents in any browser, a PowerPoint file, or a PDF on Windows.

## Prerequisites

To follow this tutorial, you will need:

- **Conduit installed** on Windows, macOS or Linux. Download the latest release from the [releases page](https://github.com/runningpixels/conduit/releases).
- **A model provider set up in Conduit** with your own API key, or a local model. This tutorial uses GLM 5.3 Flash through OpenRouter. To set a provider up, follow one of the [provider guides](https://conduitllm.com/providers.html), such as [Use OpenRouter in a desktop app](https://conduitllm.com/providers/openrouter.html).
- **A PDF with a text layer**: one you can select text in. A scanned PDF needs a model that reads PDFs directly; see the questions at the end.

## Step 1 — Attaching the PDF to a chat

A deck can start from the **Slides** page, from a short description. To build it from a document, start in a chat, where you can attach files.

Choose **Chats** in the side rail, then **New chat**. Attach the PDF in one of three ways: drag it onto the chat, choose **Ask about a file** under the composer, or choose **+** in the composer and then **Attach files…**.

The file appears above the message box with its delivery under the name. GLM 5.3 Flash doesn't read PDFs directly, so the chip says **Sent as text**: Conduit takes the text out of the PDF on your computer and sends that. With a model that reads PDFs, such as Claude, GPT-5 or Gemini, it says **Sent as PDF**, and the model also sees the layout and the charts.

Then ask for the deck. Say who it is for, how long it should be and what it should end on:

```
Turn the attached annual report into a 7-slide deck for the city council's
transportation committee. Open with the year's results, show the ridership trend
and the finances, and end with the decisions we need from the committee. Keep the
text short and put the key numbers on the slides.
```

Send the message. Next, you will review the storyline.

## Step 2 — Approving the storyline

The model doesn't build slides yet. It turns the chat into a deck, opens it in **Slides**, and writes a storyline for you to check. In this step, you will review it and build the slides.

After a few seconds the Slides page opens with the chat docked on the right. The header shows the deck title, a **Made from your chat** label with **Undo**, the **Theme**, and the steps **Story**, **Slides**, **Polish** and **Present**. Above the storyline, **Assumed** states what the model took for granted: here, the audience, the goal of securing three approvals, and a ten-minute briefing.

Under **Storyline** is one line per slide. Each line is what that slide says, not just its topic:

This is the cheapest moment to change the deck. Click a line to rewrite it, use the arrows to reorder lines, **×** to remove one, or **Add line** for another slide. If the story is wrong, say so in the chat instead and the model drafts a new storyline.

Choose a **Theme**: **Ink**, light text on a dark background, or **Paper**, dark on light. This tutorial keeps Ink. Then choose **Build slides**.

The model builds the slides in order and reports each one in the chat. After each change, Conduit checks the layout of the slides that changed, and if text is cut off, overlaps or doesn't fit, the model is told exactly where and asked to fix it. The seven slides took a minute or two.

Next, you will look at what it built.

## Step 3 — Reviewing the deck

In this step, you will check the slides against the report.

The deck opens on the **Polish** step: the slide strip on the left, the selected slide in the middle, and the chat on the right with a summary of what was built. Select a slide in the strip to see it.

The model chose a layout for each slide from the content. The ridership trend became a bar chart, the network and rider figures became rows of large numbers, the finances became two columns, and the asks became a short list. Every number came from the report: 1.42 million trips, 112 stations, 94.1% bike availability, a $998,000 surplus.

Check the numbers anyway. With a model that gets the text of a PDF, a table that came out of the PDF in the wrong order can turn into a wrong number. Compare each figure with the source, especially those from tables.

**Note:** If a build stops partway, which can happen when a model's reply is cut off, the chat says so and continues on its own with the slides that are missing. If anything is still missing, ask for it by number in the chat.

Next, you will edit the words.

## Step 4 — Editing the words in the Script view

Changing a few words shouldn't need a model. In this step, you will edit text directly.

In the panel on the right, choose **Script**. Every slide's text is shown as one document, in order: the slide number and its layout, then each piece of text on it, then its **Notes**, the speaker notes for presenting.

Click any line and type to change it. A line you have edited is marked **You wrote this. The AI keeps it.**, and later changes made from the chat leave it as you wrote it.

The search button opens **Find** and **Replace with**, with **Match case** and **Whole word**, to change a term on every slide at once. **Copy as Markdown** copies the whole script, which is useful for a handout or an email to the people who will be in the room.

Next, you will ask for a new slide.

## Step 5 — Adding a chart slide from the report

The report has a finance table the deck only summarises in words. In this step, you will ask for a chart of it.

Choose **Ask** in the right panel, and send:

```
After slide 5, add a slide with a bar chart comparing total revenue and total costs
for FY2025 and FY2026, using the figures from the report.
```

The model adds one slide in the right place and leaves the others alone:

The four values match the report's table exactly. Because the deck is built from slides rather than regenerated as a whole, you can keep asking this way: "make slide 3 shorter", "swap slides 4 and 5", "add speaker notes to every slide". Each turn that changes the deck is saved as a version. To go back, choose **History** and **Restore** on an earlier one.

Next, you will present and export the deck.

## Step 6 — Presenting and exporting

In this step, you will present the deck and save a copy to share.

Choose **Present** to start the slide show from the first slide, or **From this slide** to start from the one on the stage. F5 and Shift+F5 do the same. During the show, **Presenter view** opens a second window with your notes, the next slide and a timer, and **Black screen** blanks the room's screen while you talk.

To share the deck, choose **Export**:

- **HTML file — presents in any browser** saves the whole deck as one file. Anyone can open it and present it with the arrow keys, without Conduit.
- **PowerPoint — slides you can edit** saves a.pptx file on every platform. It matches the deck as Conduit draws it, with editable text, native charts and speaker notes.
- **PDF** saves one page per slide, for email or printing. PDF export is available on Windows for now; on macOS and Linux, export the HTML file and print it to PDF from a browser.

## Going further: other ways to start a deck

The same studio works without a PDF:

- **From a description.** On the **Slides** page, describe the deck under **What's the story?**, or pick a starting point such as **Quarterly update** or **Proposal with an ask**, then **Start deck**.
- **From several files.** Attach a report, a spreadsheet saved as CSV and a Word document to the same chat, and ask for one deck from all of them.
- **From your document collections.** Attach a collection to the chat with **+** and then **Documents…**, and the model can search it while it writes; see [How to chat with your PDFs and get cited answers](https://conduitllm.com/chat-with-pdfs.html).

## Questions

**What does it cost to make a deck this way?**

Conduit is free. The model calls are billed by your provider, to your own key. In this tutorial, with GLM 5.3 Flash through OpenRouter, the storyline, the eight slides and one change together cost $0.0035, about a third of a cent. A larger model costs more per slide; the cost of each chat shows under the composer.

**Can it read a scanned PDF?**

Only with a model that reads PDFs directly. Conduit sends the PDF itself to models that accept it (Claude, recent GPT and Gemini models, and OpenRouter models that list file input), and the chip under the file says **Sent as PDF**. Every other model gets the text, taken from the PDF on your computer, and the chip says **Sent as text**. A scanned PDF has no text layer, so a text-only model gets nothing from it.

**Can I export the deck to PowerPoint?**

Yes. **Export** has a **PowerPoint — slides you can edit** option on every platform. The.pptx file matches the deck as Conduit draws it, with editable text, native charts and speaker notes. Export also saves a single HTML file that presents in any browser, or, on Windows, a PDF with one page per slide. You can also present it from Conduit with **Present**, with a presenter view and speaker notes.

**Does my PDF leave my computer?**

It goes to the model provider you chat with, like any message: the extracted text, or the PDF itself for models that read PDFs. Conduit has no server of its own in between. To keep the document on your computer, use a local model through Ollama or LM Studio.

**How long can the PDF be?**

Conduit sends up to 150,000 characters of text per document, about 60 pages of dense prose, and up to 300,000 per message across all attachments; longer text is cut off. A PDF sent as a PDF can be up to 20 MB. For a longer report, attach the chapters you want on the slides.

## Conclusion

In this tutorial, you attached a PDF report to a chat, approved a seven-line storyline, built the deck, edited its words in the Script view, added a chart slide from the report's finance table and exported the deck to share.

The pattern is the same for any document you need to present: agree on the story first, let the model do the layout, then fix the words yourself. To go further, explore these guides:

- [How to chat with your PDFs and get cited answers](https://conduitllm.com/chat-with-pdfs.html), to ask questions across a set of documents.
- [How to do deep research with Conduit and Exa](https://conduitllm.com/research-with-exa.html), to gather the material for a deck from the web.
- [The provider guides](https://conduitllm.com/providers.html), to choose the model that builds your decks.

[Download Conduit](https://conduitllm.com/#get)

Conduit details match the latest release and are checkable in [the documentation](https://conduitllm.com/docs.html) and [the source](https://github.com/runningpixels/conduit). Found an error? [Tell us](https://github.com/runningpixels/conduit/discussions) and we will fix it.

## Related

- [How to set up MCP servers on Windows](https://conduitllm.com/mcp-servers-windows.html) — Set up MCP servers on Windows: install Node.js and uv, add local servers with npx or uvx, add remote ones by URL or from the registry, fix common errors.
- [How to chat with your PDFs and get cited answers](https://conduitllm.com/chat-with-pdfs.html) — Ask questions across several PDFs at once and see which document and passage each answer came from: build a collection, attach it, check sources.
- [How to run AI agents side by side in Conduit](https://conduitllm.com/run-ai-agents.html) — Define AI agents once, each with its own model and tools, then hand them work from chat, watch them work side by side, have their work reviewed and run them on a schedule.
- [How to do deep research with Conduit and Exa](https://conduitllm.com/research-with-exa.html) — Turn on Exa search, write a research brief and read a cited report in Conduit, with solar cell research as the worked example. No Exa key needed.
- [How to automate tasks with Conduit Workflows](https://conduitllm.com/automate-with-workflows.html) — Build a Conduit workflow that watches a project's releases, summarizes each with a low-cost model, saves the summary and alerts you only to breaking changes.
