muse creative skills
Muse operates a virtual character Instagram channel
How to add it to Muse
Copy the skill's SKILL.md (from the copy box or the source) and paste it into a Muse chat, asking Muse to follow it. This lasts for that conversation; saving skills isn't documented by Meta. In Muse Code, install it with `muse skills install` per the README.
New to Muse? Join at muse.ai/join and use invite code Referral: using this code gives the site owner credits.
There's no official Muse install link, so musebot never makes one up. Skill For the Muse app, web or WhatsApp.
Skill text to copy
---
name: "character-channel-ops"
description: "Operate a virtual-character Instagram channel (fictional persona): launch setup, character consistency, content cadence, caption/hashtag playbook, growth, monetization, and launch-blocking compliance (Meta AI labels, Korea FTC rules). Trigger: running/growing/posting/monetizing a virtual-character channel. JP caption drafting: jp-sns-ops."
metadata: { "includeInPrompt": true }
---
# Character Channel Ops (character-channel-ops)
Narrow-scoped skill for operating a fictional-persona Instagram channel.
Pilot character (example): a 21-year-old Japanese woman (university student,
photoreal style, 168cm), pillars: student/campus life 35 / fashion 40 /
Tokyo life 25.
Full research playbook: keep your own research notes (27 sources in the original) alongside the skill.
Benchmark data: keep your own AI-influencer benchmark alongside the skill.
Character bible: your character's bible file (appearance + non-visual traits).
Reference images: your character's reference folder.
Trigger ONLY when the user asks about the character channel: content planning,
posting, growth, monetization, or compliance for the virtual persona.
Do not activate for general Instagram questions or the Naver blog goal.
## Preconditions
- Never publish, schedule, or log in to Instagram on the user's behalf without
explicit authorization for that exact action. Drafting and planning need no
special approval; anything that touches the live account does.
- Never present the character as a real human. Never write fake firsthand
testimonials ("I tried this product and my skin cleared") — Korea FTC treats
them as false advertising even WITH disclosure.
## Step 1 — Compliance first (launch blockers, not afterthoughts)
1. Profile-level "AI-generated" label must be ON from day one (Meta policy
2026-08-31). Unlabeled AI-persona profiles are excluded from
Reels/Explore recommendations to non-followers. Labeled = no reach penalty.
2. Per-post "AI info"/"Made with AI" toggle ON for every AI-generated or
AI-edited post, reel, and story. Caption text saying "made with AI" does
NOT count as disclosure.
3. Korea FTC (2026-06-01): in any paid/recommendation content on Instagram,
keep the text "가상인물" near the character, continuously visible while
she appears. For text media: put it in the title or lead.
4. Korea FTC back-ad rules: ANY compensation (cash, product, coupon, affiliate
commission) → Korean-language `광고`/`협찬` at the BEGINNING of the post,
adjacent to the recommendation. `#ad`, buried hashtags, or "체험단" alone
do not satisfy this.
Pre-publish checklist: profile AI label on → per-post AI toggle on →
`광고`/`협찬` at post start if compensated → `가상인물` overlay on paid video
→ no fake experience claims.
## Step 2 — Character consistency (the core technical work)
Consistency ladder, weakest to strongest:
1. Fixed seed — cheap, breaks easily.
2. Reference anchoring (`--cref`/`--sref`/IP-Adapter) — good for series work.
3. Character sheet (10+ angles: front, 3/4, profile, close/medium/full body,
smile/neutral/pensive, signature pose) — the reference bank.
4. LoRA (rank 16–32, 20–50 images, rare trigger token) — for hundreds of
images. **Reference-based anchoring is the 2026 standard for photoreal
characters like the example persona**; LoRA stays the right call for stylized
(non-photoreal) characters.
Practical order: lock one hero shot → build the 10-angle sheet with
`--cref {hero} --cw 50` (face features only) → train LoRA, captioning
everything EXCEPT identity so the token absorbs identity → test 50–100
images, require 80%+ same-person recognition → reject drifted images from
the bank, never feed them back in. Lock style too: fixed `--sref` palette
and lighting language in every prompt.
## Step 3 — Weekly cadence (after the 21-day warmup)
- Reels 50–60% (3–4/week, discovery), carousel 20–25% (1–2/week, saves),
single photo 10–15% (1/week, world-building), stories daily (polls/Q&A).
- Reels: 7–15s sweet spot, hook in the first 3 seconds (50–70% drop off
otherwise), captions ON (muted viewers), cut every 1.5–3s, loopable edit,
1080×1920 minimum, no TikTok/CapCut watermarks.
- Ranking signals in order: watch time (70–90% completion) > DM shares >
saves > early engagement. Design every reel to be worth DMing to a friend.
- Posting times KST: weekdays 11–13 / 19–21, weekends 9–11; re-personalize
from Insights after 1–2 weeks. Never post at exactly the same minute daily;
rotate the 3–5 hashtag set per post (hashtag cap is 5 since Dec 2025:
2–3 niche <50k posts + 1–2 mid + 0–1 broad).
- Caption structure: hook in first 125 chars → story/value → ONE CTA
(save / tag a friend / comment / follow-for-series). 80% value,
20% promo.
### Hook-first edit rules (adopted 2026-10-23)
- Strongest frame at 0:00 — no fade-ins, logos, or slow starts. The end
frame visually rhymes with the start frame so the loop is seamless.
- Flop recovery: if a reel underperforms, recut ONLY the first 2 seconds,
swap the cover frame, and repost as a new reel — one variable at a time,
so the result teaches whether the hook or the body failed.
### Trial Reels gate (adopted 2026-10-23)
- New formats and unproven hooks debut as **Trial Reels** (shown to
non-followers only, off the main grid). Judge after 24–48h on completion
rate and DM shares against the channel median: winners graduate to a
regular post, losers get the 2-second recut or are dropped.
- Proven series formats may skip the trial only by explicit user decision.
### Weekly trend slot (adopted 2026-10-23)
- One slot per week is a current text-overlay trend format, selected with
`ig-outlier-research`. Prefer original-audio formats (business-account
safe); the format must map to a content pillar; log format + outcome in
the weekly review.
- Starter candidates (Oct 2026): "Not all angels are in heaven…",
"Money can't buy happiness but…", "Some people call me…".
### Handwritten h
channels/character-channel-ops/SKILL.md from the repo, MIT licence. Read it before pasting. Muse asks before it acts.
What the source says
Image & video production skills for AI agents: phone-shot realism, pose references, editorial posters, vertical video prompters, ad pipelines
Quoted from the source, by nanbada. Licence: MIT.
Where we found it
- GitHub repository ↗ 2026-10-09