The full style registry (hardcoded + Vertex AI-discovered), candidate tags
GPT-2 has harvested and not yet promoted, and the content-policy dead-letter dataset —
which (subject, style) pairs Vertex AI's own IP/safety filters have permanently blocked, kept
as data, never retried. Promotion itself happens on the box: emily
promptoverse promote <label> [--rare].
Every style emily promptoverse add can draw from — hardcoded at build time, or discovered/promoted since. rare styles are excluded from selection most of the time, on purpose.
Loading…
Parsed from emily promptoverse brainstorm completions, not yet promoted into a real style unless marked.
Loading…
(subject, style) pairs Vertex AI permanently declined to generate — never retried, kept only to spot patterns (e.g. which subjects carry IP sensitivity across which styles).
Loading…