Documentation

How FreeTextToCAD Works

Everything you need to get from text to 3D model

Quick Start

Using FreeTextToCAD takes about 30 seconds to get your first model:

Open the CAD Generator
Type a description of your part with dimensions (e.g., 40×20×5mm bracket with a 4mm hole in the center)
Click Generate 3D Model or press ⌘ + Enter
Wait 5 to 15 seconds for the AI to generate and render your model
Click Looks good to accept, or give feedback to refine
Click Export STL / OBJ to download

Writing Good Prompts

The quality of your model depends heavily on your prompt. Here are the key principles:

Always include dimensions

The AI defaults to generic sizes when you don't specify. Always give dimensions in millimeters:

✅ A 60×40×8mm rectangular plate with four 3mm mounting holes, one in each corner 5mm from the edge
❌ A rectangular plate with mounting holes

Name the shape type

Start with the geometric primitive: box, cylinder, bracket, tube, L-bracket, wedge, ring, etc. This helps the AI choose the right modeling approach.

Describe features in build order

Think like a machinist: describe the base shape first, then what gets cut away or added:

Start with a 50mm cylinder, 10mm tall. Cut a 30mm cylinder through the center to make it hollow. Add a 3mm flange around the top edge.

Everything builds in your browser. The first model you make downloads the modeling engine once (~60 MB), then it's cached for instant builds after that.

After it builds

Once a model appears, you can:

Suggest a change: describe what to adjust (e.g., "make the hole bigger", "add a fillet to the top edges", "5mm taller") and the model updates in place.
Export: download the model as STL or OBJ.
Generate a new model: start over with a fresh prompt.

Exporting Your Model

Click Export STL / OBJ once your model looks right. You can export as:

STL (binary): Most common for 3D printing. Works with Cura, PrusaSlicer, Bambu Studio, etc.
OBJ: Better for CAD software and renderers. Includes face normals.

STEP / IGES / BREP export is not currently supported; exports are mesh-based (STL / OBJ). If you need to take a model into Fusion 360, SolidWorks, or another parametric tool today, import the STL and rebuild from there.

Rate Limits

The AI models behind the generator are on free plans with a shared daily allowance, so the limits are built around models rather than individual requests. One model is not one AI call: planning, writing the code, repairing it, and reviewing the result typically add up to six to ten calls, and a difficult part can use more.

12 models per day per device, resetting at midnight UTC
A cooldown after each model, priced at about 2 minutes per AI call it used — so a typical model costs a 12–20 minute wait, and an unusually hard one costs more
Up to 40 AI calls per model, which is generous enough that hard parts get several repair attempts before giving up

The trade is deliberate: a longer wait between models buys each one more attempts to come out right. The generator shows a live countdown when a cooldown is running, and tells you how long to wait if you hit any other limit. Networks with a lot of users may also hit a shared per-network daily cap.

Supported Model Types

FreeTextToCAD works best for:

Mechanical parts: brackets, mounts, enclosures, adapters, clips
Simple consumer objects: stands, holders, organizers
Geometric shapes with extrusions, booleans, and holes
Rotationally symmetric parts (vases, cups, pulleys)
Signs, nameplates, and badges with raised or engraved lettering

It is less suitable for:

Complex assemblies with many moving parts
Very high-detail surface geometry
Organic, freeform shapes such as characters and sculptures — those go through Photo to 3D Print instead

Lettering uses a built-in engraving font covering A–Z, 0–9 and - _ . , : / + = # ! ?. Lowercase is cut as capitals. Accented and non-Latin characters are not in the font, so ask for the blank plate and add those in a slicer.

Photo to 3D Print

Some shapes can't be described in millimeters, because the shape is the picture. Photo to 3D Print is a separate tool for those, and it uses no AI at all — your image is converted to geometry directly in the browser, which means no rate limit, no cooldown, and no waiting on a model. It makes four things:

Lithophane: a backlit panel where image brightness becomes thickness.
Relief plaque: a raised carving of the image on a flat base.
Silhouette keychain: the outline of the subject, extruded, with a keyring hole.
Cookie cutter: the outline as a cutting blade with a handle rim.

Your photo stays on your device unless you tick the box offering it to us for improving the tool. High-contrast images with a clear subject work best.

Troubleshooting

My model looks wrong

Use the Give feedback button and describe what's wrong specifically: "The hole is too small", "The walls are too thin", "The proportions look off". Typed feedback is processed with context from your current model.

Generation fails repeatedly

Try simplifying your prompt. Remove fillets and rounds, reduce the number of features, and be more explicit about dimensions. Complex multi-part assemblies are hard for the AI to get right on the first try.

Export produces empty/corrupt file

This can happen after a failed generation. Try regenerating with a slightly different prompt. If it persists, contact us at mattpeshapps@gmail.com.