Runway Unlimited Hack, Seedance Prompts, Codex vs Claude — AI Daily May 24
263 messages · 60 active members
Overview
Topics
@instanetworks revealed a direct Discord contact at Runway (ixitimmyixi) who can manually enable the unlimited option on new accounts since it's no longer offered by default. With Veo3 throwing constant errors and Omni capped at 16/day via useapi, @swh800 and @arielletolome are migrating workflows to kie.ai, which reportedly handles Seedance 2 prompts better.
@KtargetMedia shared a full storyboard-based Seedance template with scene-by-scene voiceover segments, but caption/subtitle/music negatives don't always stick. @instanetworks reports only 1/15 generations have the issue with similar prompts, suggesting the model is sensitive to scene description phrasing.
@HabitsChangeU pointed to Terminal Bench 2.0 showing Claude Code ranked 51st vs Codex at 7th, challenging community assumptions. Caveat: Claude Code with Opus 4.7 isn't on the leaderboard yet (only 4.6). @MiamiMega's tiered routing config (Sonnet default, Haiku auxiliary, Codex delegation, with fallback) emerged as a pattern after he burned $197 leaving Opus as the default.
@mb29266 detailed his uploader stack — txt-file ads with page library auto-assignment, folder-based concept/lander structure that Claude can browse for instant iterations on winners. He runs 100% bid-capped 50-ad adsets that spend within 30 mins if they're going to work, doing several hundred iterations per day.
@startropics flagged Google's IO 2026 announcement of an agentic search layer, called their biggest upgrade in 25 years. @ecom2023 argued this favors brands with first-party data and strong authority since agents will prefer trusted sources. CPCs likely rise but with higher-intent clicks and better conversion probability.
Key Takeaways
- Runway unlimited plans can still be enabled on new accounts by DMing ixitimmyixi on Discord with your account email after signup.
- Claude Code ranks 51st on Terminal Bench 2.0 while Codex sits at 7th — though Opus 4.7 isn't yet benchmarked, the gap is challenging default harness choices.
- Always explicitly set model tiers in agent configs: Opus on default + auxiliary tasks can burn $200/day on housekeeping loops. Use Sonnet default, Haiku auxiliary, Codex delegation.
- Bid-capped 50-ad adsets on Meta spend within 30 mins if they're going to convert — a fast signal for which creative concepts deserve iteration.
- With Veo3 and Omni reliability degrading, kie.ai is emerging as the community's preferred Seedance 2 provider for ad video automation.
Hot Threads
Runway unlimited backdoor + Seedance caption issues
Claude Code vs Codex harness rankings on Terminal Bench
500 ads/day bid-cap workflow with Claude-driven iteration