>_ chinchilla.marketing Start a project
← all video builds ~/builds $ ./caption-clip-pack.md --pack

Clips, captioned on the moment.

A caption system for streamer and live-clip pages. It leads with the defining moment or reaction, grounds every line in evidence (frames, on-screen text, transcript), and keeps it under 220 characters.

live clipsready-to-pasteanti-slop rulesfree to unlock
caption-clip-pack.md — what's inside · eight packs · one file
The system prompt
Evidence-grounded, concrete, no generic hype.
Inputs
Streamer, stream title, evidence block.
Output contract
1-3 sentences, under 220 chars, no contractions.
Source-language rewrite
One clean English headline from a foreign caption.
Voice knobs
Character cap, emoji cap, evidence strictness.
Universal rules
No pipeline disclosure, no filler hype.
Unlock free — 12s sponsor view → See the categories
what's in the pack

Lead with the moment.

The reaction first, the scenery never. Every claim is tied to something the evidence actually shows.

01

Evidence-grounded

Frames + on-screen text + transcript; nothing invented.

02

No contractions

A distinct, clean register for clip pages.

03

Under 220 chars

Fits the feed without truncation.

how it works

From file to on-voice in four steps.

No code, no API keys — just a .md you paste into the agent you already use.

01 drop in
Add the caption-clip-pack.md to your agent's instructions.
02 point
Point it at the page you are posting to.
03 feed
Give it the inputs this pack asks for.
04 tune
Paste 3-5 real captions and run the voice-tuning loop until it locks.
what's inside

The .md caption pack.

You're buying a downloadable .md pack written for AI agents (and humans). It carries the complete caption system — system prompt, inputs, output contract, and voice knobs — plus the universal anti-slop rules.

caption-clip-pack.md
the downloadable pack — system prompt, inputs, output contract, voice knobs (this is the file you buy)
prompt + rules
ready-to-paste system prompt plus the universal anti-slop rules
format
single Markdown (.md) file — plain text every AI agent can read
contents
one complete caption system: evidence prompt, inputs, output contract, source-language rewrite, voice knobs
models
written for any capable LLM agent (DeepSeek, GPT, Claude, Gemini, local)
license
free to unlock, reuse forever; no subscription, no per-post credits
compare & contrast

Chinchilla Caption Agent Pack vs. ad-hoc prompting.

Most people retype the same caption instructions into a chat box every single day, and the voice drifts each time. This pack is a system: eight structured prompt sets with input contracts, output contracts, and the anti-slop rules that keep a page sounding human.

dimension
caption agent pack
ad-hoc prompting
What you get
8 structured systems — one per page type, each with an input + output contract.
One chat thread you re-explain from scratch every post.
Consistency
Same voice every run. The rules and structure are fixed in the prompt.
Tone drifts depending on the day, the model, and your mood.
Voice matching
Example-seeded tuning loop — paste 3–5 real captions and the voice locks in.
You manually re-describe the voice and hope it sticks.
Anti-slop
Banned-phrase list + no-disclosure rules baked into every category.
Whatever the model defaults to — usually hype and emoji walls.
Coverage
8 page types: hip-hop, clips, comedy, fandom, fitness, local business, alerts, ad copy.
One generic caption style stretched across every account.
Cost model
Free to unlock, reuse forever.
Your time, on every post, for as long as you post.
Setup
Paste one .md into your agent's instructions.
Re-prompt and re-steer in chat each session.
Best for
Streamer and clip pages where every post is a specific moment and generic hype falls flat.
One-off captions where consistency does not matter.

Keep the Universal rules. The pack opens with four cross-cutting rules — never disclose the pipeline, never add unearned CTAs, concrete beats hype, match the page's real voice — so every category inherits the same discipline.

Get the Live-Clip Caption Pack.

Free download. Unlock caption-clip-pack.md — the prompt that turns a stream moment into a caption that names exactly what happened.

downloadable .md for AI agents instant delivery
free no account · unlocked by a sponsor view
Unlock the .md →
12s watch time — scaled to this build's render workload (25 units), capped at 60s