Workshop: Build a media agent on CPU, call image and TTS models as MCP tools
Al_Grigor · x · 2026-11-01
The author is hosting a live-coding workshop on building a generative media agent:
- The agent runs on a CPU, reads a brief, and decides which tools to call.
- A text-to-image model is deployed on a Serverless Endpoint, wrapped as generateimage(spec).
- Text-to-speech is a second tool connected via MCP.
- The same generation is also run as a batch Job over 50 records to compare latency and cost per asset against the Endpoint.
Core idea: generative media apps don't need a GPU running all the time — expensive image/speech models live on GPUs and are called as tools only when needed.
Related event: Workshop: run agents on CPU, call GPU models as tools(3 posts)→
More from coding & agent
- Berkeley/Stanford paper: agents gain up to 39% by wiping memory and restarting from best result — alexcovo_eth · 2026-11-01
- Dev shares prompt trick for rapidly iterating a physics sim engine one concept at a time — doodlestein · 2026-11-01
- FrankenSim dev shares a prompt trick: swap physics terms to boost simulation realism — doodlestein · 2026-11-01
- claude -p burns ~1.8× more subscription quota than Claude Code TUI, Magpie switches — dotey · 2026-11-01
- guizang-pages-skill: turn Markdown into polished single-file HTML pages via AI agents — op7418 · 2026-11-01
- One Pipecat pipeline, six voices: classifier-fronted handover between AI characters, one-shotted by Opus — alexcovo_eth · 2026-11-01