i taught openclaw to replace my content team. now it posts tiktoks at 3am for 2¢ a slide.
I built an OpenClaw agent called MrTok that automatically generates TikTok slideshow concepts, creates AI images using Google Gemini, adds native Proxima Nova text overlay, and auto-drafts to TikTok via Postiz — for 2 cents per slide. This is the full architecture breakdown.
🎟️ Free Live Workshop — Build Your AI Ecom Agent
Sunday, March 1st @ 5 PM EST → https://ecom.live/events/13
📌 WHAT'S COVERED
This video breaks down every layer of the system:
• Why AI images look like stock photos (and the first-person POV fix)
• The scene library system: training your agent on real viral slides…
Watch on YouTube ↗
(saves to browser)
Chapters (14)
Intro: what MrTok is and why I built it
1:12
The output: full slideshow the agent made (unprompted)
2:44
Starting point: Larry skill + why I upgraded to Gemini
3:57
Architecture: the full pipeline explained
5:03
Training data: why untrained AI looks like stock photos
6:24
Scene library: JSON breakdown of viral slideshow shots
9:19
Pinterest hack: drag image → instant JSON prompt template
11:37
Persona JSON: Dog Mom breakdown
13:54
Prompt engineering: good vs. bad (the invisible photographer problem)
16:35
Text overlay: Proxima Nova, node-canvas, Apple emoji support
20:50
Dashboard walkthrough: personas, formats, concepts
22:51
Full tech stack (Next.js, Supabase, Gemini, Postiz)
23:49
Auto-posting to TikTok with Postiz
24:44
What's next + free workshop CTA
DeepCamp AI