3D Models (Beta): Build a Model From Your PDF
Start from a PDF and ask Scholarly to build a labelled 3D model of what it explains. Refine it in chat, then export it to PowerPoint, Keynote, or video. Now in Beta.
Some documents describe things that are hard to picture from flat pages: how the parts of a pump fit together, what sits inside a cell, how the tiers of a cloud architecture stack up. Reading about it is one thing. Turning it around and seeing it is another.
Starting today, Scholarly can build a 3D model of what your document explains. It is new, and it is in Beta.
What a 3D Model Is
A 3D model is an interactive workspace in your browser, with the chat beside it. You start from one PDF, or you start empty and describe what to build.
When you start from a PDF, its first pages appear in 3D. Then you ask chat for the model: "build a labelled cutaway of the pump in this manual", or "show the organelles of the plant cell in this chapter, with leader lines".
Scholarly's AI writes the model as real three.js code. That means real geometry, physically based materials, lighting, and animation, not a handful of simple shapes. A plant cell can be cut away to show its organelles, each with a leader line and a readable label. A gear train can actually turn. When you start from a PDF, the model is grounded in what its pages say, so the labels use your document's words.
How a Build Works
- Ask chat for the model, and confirm the build. A full build uses 1 AI Creation Credit, and nothing starts until you confirm.
- The build runs for a few minutes.
- You preview the draft, then apply it or keep your current model. Nothing replaces your model until you choose.
After that, most changes are small. Ask chat to reword a label, change a colour or a size, speed up an animation, or change the background. Small edits are free, and each one appears on an approval card before it changes anything. Undo brings back the previous version.
To look at a model, orbit and zoom, switch between perspective, front, side, and top views, frame everything in view, or start a turntable.
Export It Where Your Audience Is
A model is most useful when it ends up in the deck or video your team will actually watch. On paid plans, like other Scholarly downloads, you can export:
- A 4K PNG image.
- A PNG with a transparent background, to drop onto any slide.
- An 8-second 1080p turntable video (MP4).
- A .glb 3D file that opens in PowerPoint, Keynote, Blender, and web 3D viewers.
Ways Professionals Can Use It
A nurse educator preparing an in-service. Start from the anatomy chapter or the equipment manual and ask for a labelled cutaway. Export it as a transparent PNG for the slide deck, or as a .glb so it can be turned around live in PowerPoint.
An L&D team onboarding new hires. Start from the SOP or the product spec and ask for a model of the system or equipment it describes. Export a turntable MP4 for the training video. When something changes, ask chat to update the label.
An engineer or IT architect explaining a system. Start from the spec and ask for the mechanism, the equipment, or the tiers of the cloud architecture, labelled from the document. Use it to explain how the parts relate, then export it for the design review deck.
A certification candidate studying a structure. Start from your study guide and ask for the anatomy, the system, or the architecture the exam keeps referring to, labelled in the guide's own words.
Researchers can do the same with a paper's apparatus or a molecule, and the AI 3D molecule model generator page shows how that looks.
What It Costs
There are two kinds of change, and chat tells you which one it is proposing before anything happens.
- Small edits are free. A label's wording, a colour, a size, an animation's speed, or the background. Each one appears on an approval card.
- A full build uses 1 AI Creation Credit. Turning your source into a model, or rebuilding it, is a full build. You confirm it first, it runs for a few minutes, and then you preview the draft.
Exports need a paid plan. See Plans and Limits for what your plan includes.
What Beta Means
3D models are new, and we would rather you know the edges now than find them later. Today:
- Models are explanatory. They show how something works. They are not CAD-accurate or engineering-grade.
- There is no STL or OBJ export, and no 3D printing. Exports are images, a turntable video, and a .glb file.
- There are no manual 3D editing tools. You cannot sculpt or drag parts; you change a model by asking chat.
- Check what matters. The model is built from your pages, but it can still get something wrong. Look the labels over before a model goes in front of a team.
How to Try It
- Open Scholarly and go to Home.
- Open the 3D model tab, marked Beta, or choose New → More → 3D model.
- Pick a PDF from your library, upload one, or start empty.
- In the chat beside the model, ask for what you want built, and confirm the build.
- When the draft is ready, preview it, then apply it or keep your current model.
For the full picture, see the 3D models feature page. For specific jobs, start with PDF to 3D model or the 3D training materials generator. Building training material with a team? Scholarly Teams gives you a shared source library and one bill.
