GPT-Image-2 Landers, Creative Briefs, CS Automation — AI Daily May 10
357 messages · 71 active members
Overview
Topics
@c_1media detailed his process: ideate hero sections with GPT-Image-2 at 4K, extract assets via Python background removal, then loop OpenClaw + GPT-5.5 until HTML matches pixel-to-pixel. He scrapes YouTube/Reddit for customer sentiment, researches color psychology, and prefers 24hrs on one great funnel over 24 funnels in 2hrs. Cloudflare Pages beats surge.sh for scaling hosting across many funnels.
@arielletolome argued the prompt engine is the easy part — the creative brief is where the real work lives. He recommended studying actual strategists (Nick Theriot, Dara Denny, Motion, Alessio Cordeddu, Barry Hoyt, Franky Shaw) rather than scraping winning ads into Claude. @jasonakatiff is building a creative brief system today and added 3A frameworks for audience ranking.
Operators led by @victorecom shared that they're automating ~60% of 25k monthly CS tickets using Claude Code with database access, big prompts, and CS software integrations. Complex cases still require human review, and teams are evaluating moves away from Gorgias due to rising per-ticket costs, with Commslayer's new API mentioned as an alternative.
@drcopybymatt asked what workflows survived 5 months post-OpenClaw. @justingacina: conversational media buying — agents handle 1-2% incremental optimizations that weren't worth the time before. @kennyaronson: sub-agents for advertorial production (data, copy, image, code). @romainpetitjean's thread surfaced Obsidian as the default persistent memory layer, with markdown files version-controlled and 'update'/'signoff' functions systematizing repeatable processes.
@zippi101 finally got approved after 5.5 months and 6 denials. @jasonakatiff hit the DUNS requirement (2-4 weeks) and @mj888g shared the playbook: have Claude generate an Apple approval checklist, expect 4-5 review rounds, and use expedited tickets for sub-24hr feedback. Expo + RevenueCat + Clerk emerged as the proven pre-approval stack, with PostHog + DeepSeek MCP for cheap session-recording analytics.
Key Takeaways
- GPT-Image-2 + Python BG removal + OpenClaw/GPT-5.5 iteration produces pixel-perfect landers, but a full DR funnel still needs 24hrs of human-in-the-loop work.
- The creative brief — not the video engine prompt — is the bottleneck for Veo3/Seedance; study real creative strategists rather than scraping competitor ads.
- Production CS automation is hitting ~60% deflection at 25k tickets/month using Claude Code + database + API + CS software stacks, pushing teams off Gorgias.
- Obsidian is becoming the default persistent memory layer for builders juggling OpenClaw, Hermes, and other harnesses — markdown portability beats lock-in concerns.
- Apple App Store: get DUNS first (2-4 weeks), expect 4-5 review rounds, use Claude to generate the compliance checklist, and file expedited tickets for fast feedback.
Hot Threads
GPT-Image-2 asset extraction and pixel-perfect HTML buildout pipeline
Creative briefs are the hard part of Veo3/Seedance, not the prompt engine
Automating CS at 25k tickets/month and switching off Gorgias