Week 8 Set: Template and batch

Building one reusable prompt template with real slots, then running it as a batch instead of a one-off

12 April 2027

Before the session#

Bring the pipeline diagram from week 7 and pick one generation step from it — the script step, the render step, whichever one repeats the most.

In the session#

The move this week is from prompt to template. A one-off prompt describes a video; a template describes a family of them, with slots a scheduler fills in without a human rewriting the prose each time. Take your chosen generation step and write it as a template with at least three named slots — think {saint}, {weather}, {hymn_fragment} rather than a paragraph you’d retype by hand — then fill the slots five different ways and produce (or, if compute is the bottleneck, describe in enough detail that a marker can see the variation) five distinct outputs from the one template.

Then the batching-and-culling half: if you ran this template two hundred times overnight, you would not review two hundred outputs by hand. Decide what an automatic cull looks like — a similarity threshold against a reference frame, a duration check, a keyword filter on the generated script — and state, in one paragraph, what percentage of a batch you’d expect to survive it, and why that number isn’t 100%.

Finish with the scheduling plan: name your target platforms (TikTok’s queue is widely believed among creators to reward near-daily posting; YouTube Shorts is thought to tolerate burstier upload patterns; a Discord server wants irregular drops timed to events) and state a cadence — posts per day, per platform — that a template-and-batch system like yours could actually sustain without a human touching every post.

Afterwards#

Template, cull threshold and scheduling plan are the three pieces the Slop Pipeline submission asks for in miniature. Week 9’s set is supervised time to assemble all three into the full pipeline.