| name | tiktok |
| description | Automate TikTok marketing across multiple apps using Dropspace for publishing/analytics. Use when setting up TikTok content automation, creating slideshow posts, analyzing post performance, optimizing marketing funnels, or when user mentions TikTok growth, slideshows, or social media marketing for any app. |
TikTok Marketing
Automate TikTok slideshow marketing: generate โ overlay โ publish via Dropspace โ track โ iterate.
Pipeline Architecture
All platforms share a common engine at ~/markus/. No per-platform scripts โ everything runs through shared engines with platform config from platforms.js.
Data: ~/markus/apps/{app}/tiktok/ (strategy.json, posts.json, failures.json, pending-batches.json, posts/)
App config: ~/markus/apps/{app}/app.json (single source of truth for app identity, integrations, platform settings)
Key Scripts (all in ~/markus/)
node engines/create-visual-post-engine.js --app <APP> --platform tiktok --next --schedule "2026-03-04T12:00:00Z"
node engines/self-improve-engine.js --app <APP> --platform tiktok --days 14
echo '{"posts":[...],"notes":"..."}' | node scripts/add-posts.js --app <APP> --platform tiktok
TikTok Details
- Format:
story-slideshow (default) โ 5 content slides + engine CTA (1024ร1536 portrait).
- 3 posts/day at 8AM, 1PM, 6PM ET
- Image model: gpt-image-1.5 (never gpt-image-1)
- Privacy:
SELF_ONLY โ posts land as drafts in your TikTok inbox. He adds trending sound, changes privacy to public.
- Music:
auto_add_music: true as fallback. Manual trending sound = ~100x more views.
The Larry Slide Formula (5 content + engine CTA)
| Slide | Purpose | Text Style |
|---|
| 1 | HOOK โ stop the scroll | Full hook text, relatable problem |
| 2 | PROBLEM โ amplify pain | Build tension |
| 3 | DISCOVERY โ turning point | "So I tried this..." |
| 4 | TRANSFORMATION 1 โ first result | "Wait... this actually works?" |
| 5 | TRANSFORMATION 2 โ escalate | "Okay I'm obsessed" |
Note: Slide 6 (CTA) is engine-appended. The LLM generates 5 content slides only.
Text Overlays
Sizing:
- Short text (โค5 words): 7.5% of image width
- Medium (โค12 words): 6.5%
- Long (13+): 5.0%
- Outline: 15% of font size
Positioning:
- Centered at ~28% from top (safe zone)
- Top 10% hidden by TikTok status bar
- Bottom 20% hidden by TikTok UI
- Max width: 75% of image
Content:
- REACTIONS not labels: "Wait... this is nice??" not "Modern minimalist"
- 4-6 words per line
- 3-4 lines per slide ideal
- No emoji (canvas can't render them)
- Full hook on slide 1 โ never split across slides
Content Rules
Tone
- Most emotional of all platforms โ TikTok rewards vulnerability and relatability
- Storytelling > information โ "i literally cried" beats "here are 5 tips"
- All lowercase โ TikTok native voice
- Emoji encouraged โ ๐ญ ๐ซ ๐ are TikTok-native
- No corporate voice โ instant scroll-past
Caption Formula
[hook matching slide 1] ๐ญ [2-3 sentences of relatable struggle].
So I found [APP NAME] that [what it does in one sentence] โ
you just [simple action] and it [result].
I tried [thing 1] and [thing 2] and honestly?? [emotional reaction].
[funny/relatable closer]
#hashtag1 #hashtag2 #hashtag3 #hashtag4 #fyp
Rules:
- Storytelling, not feature lists. Long captions get 3x more views.
- Mention app naturally โ never "Download now!"
- Max 5 hashtags
- Conversational tone, like texting a friend
Hook Formulas
Tier 1: Person + Conflict โ Result (consistently 50K+)
- "[Person] said [doubt/insult] so I showed them [result]"
- "I showed my [person] what [app] can do and they couldn't believe it"
Tier 2: Relatable Pain โ Discovery
- "I was spending [X hours] on [task] until I found this"
- "POV: you have [aspiration] but [constraint]"
Tier 3: Curiosity / Challenge
- "I tried to [challenge] using only AI"
- "Can AI actually [thing people doubt]?"
What DOESN'T work: Feature lists, self-focused complaints without conflict, fear/insecurity hooks.
Post Blueprint Structure
When generating posts for the queue via add-posts.js, the blueprint structure depends on the format:
Visual formats (story-slideshow)
story-slideshow uses sceneAnchor + slideMoods (locked architecture โ see formats.js and formats.js story-slideshow entry):
{
"format": "story-slideshow",
"text": "Hook text",
"slideTexts": ["Slide 1 text", "Slide 2", "Slide 3", "Slide 4", "Slide 5"],
"sceneAnchor": "Detailed scene (150-300 words): room, furniture, person, camera. Shared across all slides.",
"slideMoods": ["mood delta 1 (30-60 words)", "mood 2", "mood 3", "mood 4", "mood 5"],
"caption": "Storytelling caption with #hashtags"
}
Video formats (ugc-reaction)
{
"text": "Hook text",
"videoPrompt": "6-second video description...",
"caption": "Caption"
}
Engine auto-appends CTA slide for formats with ctaSlide: true.
Self-Improvement Thresholds
- Tier 1 (double down): 50,000+ avg views
- Tier 2 (keep): 10,000+ avg views
- Drop threshold: 5,000 avg views
- Individual hook drop: <1,000 views
- Winner for variations: 10,000+ views
Note: The pipeline now uses engagement rate as the primary optimization signal. Raw view thresholds are secondary guidance.
Diagnostic Framework
| Views | Conversions | Diagnosis | Action |
|---|
| ๐ข High | ๐ข High | SCALE | 3 variations of winning hook |
| ๐ข High | ๐ด Low | FIX CTA | Rotate CTAs, check landing page |
| ๐ด Low | ๐ข High | FIX HOOKS | Test different hooks, keep CTA |
| ๐ด Low | ๐ด Low | FULL RESET | New format, new audience angle |
Attribution
First-touch attribution (if configured) captures referral source:
- UTM: `
- TikTok caveat: single "link in bio" โ per-post attribution uses signup timing vs post timing (72h window)
Common Mistakes
| Mistake | Fix |
|---|
| Landscape images (1536x1024) | Portrait always (1024x1536) |
| Font too small (5%) | 6.5% of width minimum |
| Text at top/bottom of image | Position at 28% from top |
| Different scenes per slide | Lock architecture, only change style |
| Labels instead of reactions | "Wait this is nice??" not "Modern style" |
| gpt-image-1 instead of 1.5 | Always gpt-image-1.5 |
| Generic image prompts | Be obsessively specific |
