A 5-second talking-head loop in 6GB VRAM: local graph, hosted lip-sync API
Extreme-Shock1930 · reddit · 2026-10-01
Pushing back against 40-node talking-head workflows, the author built the smallest possible graph: load video, load audio, plus a sync custom node set (video in, audio in, api key, generate, output, save) that turns a 5-second clip and a voice recording into a synced video.
Loading and preview run on a 3060 with barely any VRAM movement; the actual lip-sync is a hosted API call, so local GPU specs don't matter — you pay per second of output instead of waiting on VRAM.
Open questions: where to cache results so a graph re-run doesn't re-bill the same segment, and whether to keep the API key in the graph JSON or in env. He asks how others split local vs hosted stages.
More from coding & agent
- Why Shopify dropped React Native: AI agents made native the future again — davemccollough · 2026-10-01
- ClawDaddy MCP lets AI agents buy domains via Stripe and manage DNS end-to-end — modelcontextprotocol · 2026-10-01
- MetricDuck MCP serves SEC financial data to AI agents with filing-level traceability — modelcontextprotocol · 2026-10-01
- A 1-cent verifier catches 61% of AI agents falsely claiming task completion, paper finds — alex_verem · 2026-10-01
- Giving Claude full PC control built a playable SCP-096 horror game in 3 hours — imjustnewatai · 2026-10-01
- OpenAI dots experiment: a handoff-note template to test agent memory across sessions — Kitchen-Jicama8715 · 2026-10-01