Dev releases polished 3-week ComfyUI workflow for MiniMax H3 with multi-reference pipeline
altoiddealer · reddit · 2026-09-19
Reddit user altoiddealer shared a ComfyUI workflow for MiniMax H3, built from scratch over 2-3 weeks with an emphasis on structure and extensibility.
Key features:
- Supports up to 6 images, 2 videos, and 3 audio inputs; any can serve as reference, guide, or both — flexibility the author says is unmatched by other workflows
- A 1-4 step LLM prompt-enhancement pipeline (Ref2VA): prompt expansion → reference analysis (Reference Map) → Creative Director blueprint → Prompt Compiler for the H3 prompt structure, with human-editable LLM outputs between stages
- A one-step FL2VA enhancer that auto-detects mode (T2VA/I2VA/FL2VA/L2VA) from image state
- Seamless video continuation and forced audio options (borrowed from the Seed Hunter workflow), plus easy LoRA/speedup expansion
- Requires up-to-date ComfyUI frontend (<v1.50.5) and custom nodes including ComfyUI-Easy-Use and KJNodes
More from coding & agent
- Hindsight ships 0.8B model for agent memory decisions: 2.5x cheaper, 1.9x faster — victor_explore · 2026-11-04
- Human Oversight Is 70-75% of AI Agent Costs in Banking, McKinsey Finds — mikeflache · 2026-11-04
- Over 80% of Major Financial Institutions Now Run Agentic AI in Production — mikeflache · 2026-11-04
- Mnemos MCP nears launch after shared memory layer fix — RileyRalmuto · 2026-11-04
- Moving mobile dev to remote machines makes it multiplayer across product, design, engineering — rudrank · 2026-11-04
- Kuaishou's EvoRec: self-evolving agentic recommender lifts ad revenue 4.6% — _reachsumit · 2026-11-04