Glossary
Terms you'll see in the app and docs.
| Term | Meaning |
|---|---|
| Run | A single evaluation job: one model, one or more dimensions, and a runner. A run goes through states like pending → running → completed (or failed/cancelled). |
| Runner | A GPU worker you install and register. It pulls run tasks from WorldJen, generates videos (and optionally runs local steps), and reports back. You can have multiple runners across different machines. |
| Model | A video (or world) model used in a run. Typically identified by a Hugging Face repo or similar; configured with repo URL, type, and optional auth. |
| Dimension | An evaluation axis (e.g. motion smoothness, physical mechanics, semantic adherence). Each dimension is scored per video; you select which dimensions to use when creating a run. |
| Dimension preset | A named group of dimensions (e.g. "Motion Stability", "Logic & Physics", "All Dimensions") to quickly select a standard set. |
| Prompt | The text instruction or scenario used to drive video generation for a run. A run will likely have multiple prompts; each gets its own set of generated videos. |
| Score | A numeric result for a dimension or video, usually 0–100. Higher means better alignment with what that dimension measures. |
Runner-specific
| Term | Meaning |
|---|---|
| Registration token | One-time token shown when you create a runner. Used by the worldjen register command. Expires after a short window. |
worldjen svc | Commands: install, start, stop, status, logs (and logs -f). |
App navigation
- Dashboard — List and manage runs; create new runs.
- Models — Add and manage model configurations (e.g. HF repos).
- Runners — Create runners, copy install/register commands, start/stop, view status.
- Settings — Account and app preferences.
- Run page — For one run: summary, dimension scores, gallery of videos, logs, and options to cancel, duplicate, or export.