Summarize a PDF with AI, then illustrate it

A PDF dropped on the canvas, two Text nodes that read it in full: one summarizes it in five points, the other turns it into an image prompt, and an Image node renders it. The demo brief is included. The whole workflow is visible without an account.

Interactive demo

Real workflow, real results. Pan and zoom freely.

Use this workflow

The real workflow: a launch brief as a PDF in a Media node, wired to two Text nodes, Claude Haiku for the summary, GPT-5 Mini for the prompt, and a Seedream Image node that renders the described visual.

At a glance

What the workflow produces: 1 image in 1:1
Canvas structure: 5 nodes wired by 3 connections
Models wired in: Claude Haiku 4.5, GPT-5 Mini, Seedream 5 Pro
Cost of one full run: about 8 credits, roughly $0.10

A PDF is not an image: it is a document that language models read in full, tables included. Drop it on the canvas, it enters a Media node, and any Text node set to a model that reads files (GPT, Claude, Gemini, Kimi K3 and the others marked as such) receives it through its “Image or PDF to analyze” port.

The demo starts from a four-page launch brief. A first Text node on Claude Haiku 4.5 summarizes it in five points; a second on GPT-5 Mini turns it into an image prompt faithful to the document's visual direction; an Image node on Seedream 5 Pro renders that visual. A report becomes a visual in three nodes.

The price follows the size of the document: reserved when the file is submitted, then adjusted on the tokens actually read, the difference returning to the balance. Up to three documents per node. Minutes, product sheet, contract, course, specifications: anything that can be read can be summarized, rewritten, illustrated or turned into a carousel.

How it works

1
Drop your PDF on the canvas
A Media node receives it, like an image or a video. The file name displays, and the node tells you which port to wire it to.
2
Wire it to a Text node that reads documents
The “Image or PDF to analyze” port appears as soon as the chosen model can read a file. Write the instruction: summarize, extract, rewrite, translate, plan.
3
Generate the summary, then the prompt
Each Text node works on the whole document. The price, reserved on the file size, settles on what was actually read. The answer displays in the node.
4
Make it an image, or more
The prompt written by the second node feeds an Image node through its prompt port. Add other Image nodes for the other visual series of the brief.

Variations and use cases

The same canvas, tuned for different needs: each variation is two or three lines of prompt away.

The brief that becomes a campaign

That is the demo: summary for the team, prompt for the visual, image rendered. Add one Image node per series requested in the brief, and a Video node for the vertical formats.

The report as a carousel

Summarize a report in six points, one per slide, and chain with the carousel workflow: the text of each slide comes from the document, not from a rewrite from memory.

The manual as a video tutorial

A product manual as a PDF, a Text node that turns it into a three-step script, a Voice-over node that reads it, one Image node per step, and the Montage node that assembles. The tutorial starts from the official document.

The contract explained

Wire a contract or terms and conditions and ask for the five points that bind the signatory, or the differences with a previous version wired as a second document.

The course as revision sheets

A thirty-page course becomes one revision sheet per chapter, then quiz questions. One node per instruction, the same PDF wired to each.

The product sheet translated and illustrated

A technical sheet in English, summarized and translated by a Text node, then a product visual generated from its characteristics. The product photo workflow takes over.

Common mistakes

Picking a model that does not read files

Without the “Image or PDF to analyze” port, the document is not sent and the answer is made up. Check that the port appears before generating.

A vague instruction

“Analyze this document” gives a general text. “Five points, one per line, no introduction” gives exactly what you want, as in the demo.

A scanned PDF without text

A scan of images without a text layer reads poorly. Run it through a character recognition tool first, or export the PDF from the original document.

Sending a hundred pages for one question

The price follows the file size. For a precise question on a long document, export the useful pages as a separate PDF before dropping it on the canvas.

Recommended models

Frequently asked questions

Which models read a PDF?

Those that show the “Image or PDF to analyze” port: GPT-5 and its variants, Claude, Gemini, Kimi K3, Grok 4.6, DeepSeek V4 and the others marked as reading files. The other Text nodes ignore the document.

How much does reading a document cost?

The price follows the size of the PDF and the model's rate: reserved at launch from the file weight, then adjusted on the tokens actually read. The amount shows before you click.

Is the document kept?

It is stored in your library, like an uploaded image, and is only sent to the model when you generate. Delete it from the library whenever you want.

Can I wire several PDFs?

Up to three documents per Text node. The model receives them all and your instruction can compare them, merge them or extract a synthesis.

What PDF size is accepted?

The upload cap is the library's, and reading is bounded to a reasonable volume of text per generation. A thirty-page report goes through without a problem.

Is the summary faithful to the document?

The model reads the actual text, tables included, and does not work from memory. Still, reread the figures before publishing them, as with any assistant.

Can I ask several questions about the same PDF?

Yes: one Text node per instruction, all wired to the same Media node. Each generation is paid separately, on what it reads.

Can the PDF feed an Image node directly?

No, an Image node expects a prompt. Have that prompt written by a Text node that reads the document, as in the demo, then wire it into the Image node.

Summarize a PDF with AI, then illustrate it

Use this workflow

Free account, no card required. The workflow opens pre-filled in your canvas.