AI 3D Models · Beta

Some things are easier to understand when you can turn them around. A 3D model in Scholarly is an interactive, labelled object built from your material — a process laid out step by step, the structure of a machine or a building, a network stack, a molecule — that you orbit, animate, and drop into a deck.

You do not model anything yourself. You describe what you want in the chat beside the viewer, or point it at a PDF, and Scholarly builds it. It is in Beta, so expect it to get better and expect the occasional model that needs a second attempt.

Creating a 3D Model

Start one from Home, from the New menu, or from your library's create row. Two ways to begin:

• Start empty — the viewer opens blank and you describe the model in chat. Best when the subject is in your head rather than in a file: "Build a labelled model of the TCP/IP stack", "Model a water treatment process, step by step", "Show a caffeine molecule with labelled atoms".
• Start from a PDF — pick a PDF from your library. Its pages open in the 3D space, and the model is built from what those pages actually say. Ask "Build a 3D model of this PDF's main idea" or "Turn the process in this PDF into an animated 3D model".

Only PDFs can be used as a source. Pick another file type and Scholarly tells you so rather than failing after you press Create.

Describing What to Build

The chat sits beside the viewer and always has the model — and your source, if there is one — in context. It suggests openers when you arrive, and anything you type is an instruction about the model rather than a general question.

Good instructions name the subject and how it should read:

• "A five-stage onboarding process, left to right, numbered, with a label on each stage."
• "The layers of the Earth as a cutaway sphere, each layer labelled with its depth."
• "Animate the flow of water through the plant, slowly, in a loop."
• "Walk me through this model step by step."

You can also ask for changes to how it looks: "Make the colours easier to tell apart", "Use a lighter background for slides".

Builds and Free Edits

There are two kinds of change, and they cost differently.

A full build takes a few minutes and tells you where it is: planning, building, then checking its own work and refining it. Editing is paused while a build runs.

When it finishes you get Draft ready to review. Nothing changes until you choose:

• Preview draft — see the draft in the viewer. Nothing is saved yet.
• Apply draft — make it your model. Undo brings back the previous version.
• Back to current — keep what you had and discard the draft.

If the model changed somewhere else while a build was running, Scholarly asks before replacing your newer work rather than overwriting it silently.

A build that fails or is cancelled returns its AI Creation Credit, and chat can try again. If you have no credits left, nothing is spent — the notice tells you to add credits and press Build again. See Plans and Limits and AI Credit Packs.

Looking at the Model

The viewer has its own controls, and none of them change the model itself — they are your camera, not your work, so they are never saved and never undone.

• Drag to orbit, scroll to zoom.
• View picks a fixed angle: Perspective, Front, Side, or Top.
• Frame all fits the whole model back on screen when you have zoomed somewhere odd.
• Turntable spins the model slowly — the easiest way to present one without touching the mouse.

Keyboard shortcuts, as in any 3D editor:

Undo

Undo steps back through every change to the model, including an applied draft. It is in the header and on Cmd+Z / Ctrl+Z. It is disabled while a build is running, because a build works from the version it started on.

Exporting

Export in the header hands the model to whatever you are actually making. Exports are a paid feature.

For Keynote, use an image or turntable video. Native 3D objects in Keynote require USDA, USDC, or USDZ; Scholarly exports .glb, so that file needs conversion first.

The turntable video is recorded in your browser, so it needs a browser that can record: Chrome, Edge, or Safari. If yours cannot, Scholarly says so instead of producing a broken file.

When a Model Hits an Error

A 3D model can end up in a state that will not draw — usually after an ambitious edit. The viewer says so and offers the two things that fix it: Undo the last change, or ask chat to fix it. Chat can read the error and repair the model, and that counts as an edit rather than a new build.

Where 3D Models Live

A 3D model is an item in your library like any other. You can rename it by clicking its title, pin it, put it in a folder or a project, and find it with Cmd+K. Chat elsewhere in Scholarly can find your 3D models by name, so "open the mitosis model" works from a normal conversation.

Creating a 3D model counts as a creation for your plan's allowance in the same way a deck or a lecture does.

Tips

• Name the shape, not just the subject. "The nitrogen cycle as a labelled loop with arrows between stages" beats "the nitrogen cycle".
• Build once, then edit. Edits are free and a rebuild is a credit — so steer the model you have rather than describing it again from scratch.
• Ask for presentation colours. A model built for a screen is often too dark for a projector; "use a lighter background for slides" fixes it in one free edit.
• Export the .glb for a deck you will present live, and the turntable video for one you will send.
• Start from the PDF when the detail matters. A model grounded in the pages uses the source's own terms and labels.

Related

• Choosing What to Create — when a 3D model is the right output, and when a diagram or a video lecture is.
• AI Video Lectures — a narrated explanation rather than an object you turn around.
• Mind Maps and Flowcharts — flat diagrams, when the structure is the point and depth is not.
• Exporting Your Work — every export format across Scholarly.
• Plans and Limits — AI Creation Credits, and which exports need a paid plan.