Which path is mine?
There are two supported ways to drive NameFrame, and they produce the same artifacts. The difference is where you sit: inside the Unreal editor, or at a terminal.
Path A: no Python
Unreal Studio panel
Everything happens in a docked panel inside the Unreal editor. You click through nine pages, press Generate Dataset, and watch a progress bar. Studio runs the Python pipeline for you in the background, so you never type a command.
Choose this if you are an artist, level designer or ML engineer who wants datasets out of a scene you already have, and has no interest in the command line.
Path B: Python
Command line and SDK
You write a YAML recipe, compile it, and run commands. It scripts, it goes in CI, it runs on a machine with no editor open, and with the demo generator it can produce a dataset with no Unreal at all.
Choose this if you want reproducible runs in version control, batch generation, experiments, or a pipeline that runs unattended.
Side by side#
| Unreal Studio (no Python) | Command line (Python) | |
|---|---|---|
| Where you work | A panel in the Unreal editor | A terminal |
| Needs Unreal open | Yes, always | Only for real capture |
| Needs a GPU | Yes | Only for real capture |
| Authoring zones and classes | Buttons, viewport selection | YAML recipe |
| Reproducible in git | Via a saved recipe YAML | Natively |
| Unattended or CI runs | No | Yes |
| Batch and queues | One run at a time | Queue, control plane, workers |
| Try it with no assets | No | Yes, with --demo |
They are not exclusive#
The two paths share one file format. Studio's Save recipe button writes
the same YAML that nameframe studio-compile consumes, and Studio's Load
recipe reads a YAML you wrote by hand. A common working pattern:
Author visually
Place zones and assign classes in Studio, where you can see the level.
Save the recipe
Dataset → Save recipe writes a
studio_*.ymlyou can commit.Scale it up on the command line
Run the committed recipe with
capture-unreal, sweep seeds, queue jobs, run experiments.
A third option: let an AI drive#
NameFrame ships an MCP server, so an assistant such as Claude can read your scene, propose a scenario, dry-run it, and ask for approval before generating anything. Read-only tools are free to call. Anything that writes defaults to a dry run and needs explicit approval, which you give from a queue inside the Unreal Studio panel.