ComfyUI workflow: consistent AI characters in MiniMax H3 with just 6GB VRAM, no LoRA
cgpixel23 · reddit · 2026-10-11
A new ComfyUI workflow replaces character LoRA training with RefMod custom nodes: one reference image becomes 8 angle views via Qwen Image 2.1, packed into a 100MB character file, then used in an optimized MiniMax H3 video workflow. Tested on 6GB VRAM, no training data needed; full tutorial and workflow files are linked.
Related event: RefMod Workflow Brings Training-Free Character Consistency to MiniMax H3(3 posts)→
More from Multimodal
- Sound / Vision: open-source self-hosted app turns an idea into songs, album art, and AI music videos — LawrenceOfTheLabia · 2026-10-11
- Gausify Orbit: a web app that turns videos into Gaussian Splats in the browser — obraiadev · 2026-10-11
- Sound / Vision: open-source self-hosted app turns an idea into songs, album art, and AI music videos — LawrenceOfTheLabia · 2026-10-11
- Gausify Orbit trains Gaussian Splats from AI videos fully in-browser via WebGPU, no CUDA or uploads — obraiadev · 2026-10-11
- Claude Opus 5.5 turns Jay Chou's 'Qi Li Xiang' into Monet-style MV via skill — AlchainHust · 2026-10-11
- Midjourney style recipe: triple --sref codes with --profile and sw 50 — ciguleva · 2026-10-11