Research log · 2026-10-12 · TikTok slideshow marketing for indie apps¶
Questions¶
- What is the "TikTok slideshow" process people say drives installs, and where did it come from?
- Which numbers are real (with a source) and which are hype?
- What skills, repos and tools exist to run it, and which are worth using?
- What do TikTok's API and policies allow?
Sources read¶
| Source | Type | What we took |
|---|---|---|
| TikTok Content Posting API: photo post reference | Official docs | media_type: PHOTO, post modes, 35 images, title/description limits, rate limits, unaudited = private only, is_aigc → tiktok-photo-post-api.md |
| Integrity and Authenticity, Branded Content Policy, Commercial Content Disclosure | Official policy | Multi-account, automation, fake engagement and disclosure rules → docs/07 |
| gameplaydev Substack: Larry step-by-step | Secondary, quotes the founder | Hooks with view counts, costs, failures, setup → case study |
| Stack Junkie: Larry skill deep dive | Third party | 6-slide skeleton, decision table, warm-up practice |
| Larry on ClawHub | Skill listing | Text rules, safe zones, hook thresholds, posting defaults, file layout |
| Luminary Lane: from Larry to Lane | Vendor analysis | Where DIY agents break (drift, auth, single channel) |
| Postiz: automate TikTok slideshows, Postiz: Quran app case | Vendor | Hook-mining prompt, render specs, 3-slide case → case study |
| MrrStory: Alex Nguyen, Starter Story summary (Vix), Yahoo Finance (Candle), Building Afrika | Interviews | Other founders → notes |
| GitHub API (2026-10-12) for ~15 repos | Data | Stars, licence, last push → tools-and-repos.md |
| Vendor guides (makeviral, reelbase, socialrails, verticalflair, instacarousel) | Vendor | Only to cross-check slide counts and canvas sizes |
Findings¶
- The process went viral in Feb–Mar 2026 through Oliver Henry's "Larry" (OpenClaw agent) story and Greg Isenberg's podcast. It's a loop: research → 6 slides with a locked scene → draft via API → human adds sound and publishes → analytics + revenue → rewrite hooks/CTA.
- The hook explains most of the variance. Same pipeline: 147K–234K views for "person + doubt → showed them" hooks; under 3K for feature or self-focused hooks.
- Cost per post is tiny (~$0.50 reported), which makes many tests cheap. That, not a secret algorithm boost, is the "wild" part.
- Measure to revenue. Both the Larry loop and Vix tie content to purchases, not views.
- The API supports photo posts but not sound choice, and unaudited apps can only post privately, so everyone ends with a human publish step.
- Repos: the most-starred open slideshow generator is
athcagithub/SlideSmith(335★, licence not declared); the original Larry skill is free on ClawHub (MIT-0);jorge-dev/swipekit(MIT) is the only one focused on finding formats that work in a niche.
Not verified (don't quote as fact)¶
- "Slideshows get 5× more reach than video" and "TikTok has been pushing photo mode since early 2026": vendor blogs, no data.
- "TikTok detects Midjourney images in 0.5 s and downranks them"; "API posts from server IPs are penalised": Postiz blog, no data.
- "Warm up a new account 7–14 days": practice from the Larry skill, not a TikTok rule.
- Larry's MRR: $588, $670 and $714 appear in different places; 8M views in one week mixes X article views with TikTok.
- Best posting times; slide-1 swipe-through targets like "70%".
Output of this research¶
Docs 01–07, case studies, templates, skills social-tiktok-slideshow and social-tiktok-slideshow-loop.