Gemini Omni vs Seedance, Cursor Composer 2.5 — AI Daily May 20
416 messages · 76 active members
Overview
Topics
Google's Gemini Omni is live on kie.ai and useapi.net at $0.12-$0.60/gen, winning on text-in-video, audio quality, and editing — but Seedance 2.0 still leads on character realism and physics. Tencent's Hy3-preview also dropped and hit #1 on OpenRouter at $0.066/1M tokens. Hopes are high Omni sets the stage for Veo 4.
Composer 2.5 is drawing heavy praise — @Guto_gouvea burned 280M tokens on the $20 plan with strong results on ffmpeg, Puppeteer, and AWS debugging. Bloomberg also reported SpaceX plans to acquire Cursor 30 days post-IPO. Emerging pattern: Gemini for architecture, Composer for execution. GPT-5.5 remains the default LLM workhorse, with Grok offering no meaningful edge.
Builders converged on a workflow: Higgsfield for bulk prompt testing, kie.ai for production automation via API, Seedance through Dreamina for cheap multi-clip stitching. Debate on whether custom voicegen + lip-sync beats default model audio — voice is cheap but lip-sync friction often isn't worth it. New open-source video-analyzer skill wrapping Gemini was shared.
@sibunting is running GPT-5.5 on Slack + Opus 4.7 on Telegram + openclaw and Hermes with shared memory. @weslindquist asked for hardening tips on a 4-openclaw + Hermes + Claude Code setup — consensus: containerize and isolate after a postgres crypto-miner incident. Hardware envy: 8x RTX 6000 Pro rigs at $80-100k, and @jcartu's 4x RTX Pro 6000 running Qwen 397B locally to replace $250/day Claude spend.
Adam Young and Harrison Gevirtz pleaded guilty to assisting tech support fraud schemes, hitting DOJ and TCPAWorld. Community consensus: likely probation plus restitution, no prison. Discussion centered on how rampant the pay-per-call vertical was in that era — 'the ACA of its time.'
Key Takeaways
- Gemini Omni beats Veo 3.1 on text/audio/editing but trails Seedance on realism — wait for Veo 4 for the real test
- Cursor Composer 2.5 at $20/mo handled 280M token workloads; pair Gemini (architecture) + Composer (code)
- GPT-5.5 is the consensus default LLM — Grok offers no meaningful upgrade even for paid Supergrok users
- Local inference is real: 4x RTX Pro 6000 running Qwen 397B replaces $250/day Claude spend
- Containerize and isolate agents — a postgres exposure led to a crypto-miner compromise
Hot Threads
Ringba founders plead guilty — tech support fraud fallout
Cursor Composer 2.5 reviews and workflow vs Claude Code
Higgsfield vs kie.ai for automated AI video ad workflows