Creator Archetypes

Three kinds of talent. One consent-aware roster.

A platform that treats a consented digital twin and an invented character as the same object will eventually get someone into trouble. JimFluencer models them as distinct types with distinct obligations.

HUMAN

Real Human

A working creator using the studio purely as a production facility.

No synthetic likeness involved. The creator shoots or supplies their own imagery and uses the planner, publisher, and vault to run their calendar at volume.

Consent: Not applicable
TWIN

Digital Twin

A consented likeness of a real person, trained to scale their own image.

A LoRA is trained on a dataset the person supplied and consented to. Face-Lock keeps the resemblance stable, and the Identity Vault records likeness and voice consent with timestamps.

Consent: Required, recorded, revocable
VIRTUAL

Virtual Creator

An original character with no real-world counterpart.

Fully synthetic talent you own outright. Still requires AI disclosure on platforms that mandate synthetic-media labels, which the Trust Center checker handles.

Consent: Disclosure only

LoRA Training

A training pipeline that admits when it has died

The most common failure in self-hosted LoRA training is not an error — it is a job that says TRAINING forever because the worker crashed without telling anyone. The state machine treats that as a first-class case.

PENDINGDataset uploaded, awaiting a free worker.
DISPATCHEDClaimed by a worker, initialising CUDA.
TRAININGStepping through epochs, heartbeat every few seconds.
COMPLETEDWeights uploaded to your R2 bucket.
FAILEDWorker reported an error, or rebooted mid-run.
TIMED_OUTNo heartbeat for 120 seconds — presumed dead.

Boot-time reconciliation

When a worker starts, it announces itself. Any job still marked DISPATCHED or TRAINING against that worker must be dangling — the worker cannot have survived — so those jobs transition immediately to FAILED with a clear reason rather than hanging indefinitely.

Retry without re-uploading

Failed and timed-out jobs expose Retry Training and Reset Job actions. Your uploaded image dataset is preserved throughout, so recovering from a worker crash never costs you the twenty minutes you spent curating it.