VRAM guard trick: run a local LLM and ComfyUI video generation side by side without OOM
Sn0opY_GER · reddit · 2026-09-10
A Redditor shares a hands-on VRAM orchestration workflow using a local agent (Hermes) with ComfyUI:
- Before running minimax h3 video generation, have the agent unload the local LLM and set a "VRAM GUARD" that blocks LLM reloading during generation to avoid OOM, then reload after.
- The agent writes the script end-to-end: batch to Comfy, unload, wait for videos, and send results back via Telegram or other channels.
- Multi-workflow chaining works too: send a phone photo, edit it with FLUX.2, then feed it as the start image to the minimax h3 workflow — a clip arrives 100 seconds later.
This lets a 27B Qwen model with max context run in parallel with video generation with negligible speed loss.
More from coding & agent
- YC to host 'Make Something Agents Want' hackathon Oct 17-18 in SF as agents become new users — ycombinator · 2026-09-10
- PowerChat: an iOS client that connects cloud/local models to any MCP server — powerchat-dev · 2026-09-10
- Next AI adoption wave will be computer use: Astra already automates real office workflows, argues thread — zephyr_z9 · 2026-09-10
- Stripe Launches Agentic Treasury in Public Preview, Lets AI Agents Move Money — jeff_weinstein · 2026-09-10
- Meta's Stilla acquisition is a distribution bet, not an agent capability bet — arthaudm · 2026-09-10
- Indie dev builds 15-book multilingual kids' reading app with Fable 5.1 in 3 days — GCWebDesigner · 2026-09-10