ComfyUI MiniMax H3 Guide: Ref2V, Timeline Control, and Turbo Optimization
Fun_Walk_4965 · reddit · 2026-08-28
This advanced guide for ComfyUI MiniMax H3 focuses on improving workflow control, speed, and stability. It covers four key techniques:
- Ref2V: Uses multiple references (images, video clips, audio) to guide character identity, style, motion, and sound separately when a single starting image isn't enough.
- AddGuide: Places specific guidance at exact timestamps (e.g., frame 0, 60, 90) to control generation moments precisely along the timeline.
- Turbo LoRA: Significantly speeds up iteration by reducing sampling steps. It's recommended for testing only, as it may lose detail in fast motion or audio.
- VRAM Optimization: Addresses memory leaks by checking offloading, reducing resolution, and avoiding loading multiple checkpoints simultaneously.
The recommended strategy involves starting with basic T2V/I2V, iterating with Turbo, applying Ref2V/AddGuide for control, optimizing VRAM, and then rendering the final output.
More from Multimodal
- The math: 5 rerolls at 480p plus a 4K upscale costs $5 vs $9 at 720p — HeyAmit_ · 2026-10-15
- Dreamina workflow: generate at 480p, upscale to 4K, cut video costs 44% — HeyAmit_ · 2026-10-15
- One-prompt GPT-6 Astra writes two Palestrina-style motets in LilyPond — deanwball · 2026-10-15
- DiffSynth-Studio Adds OPD Workflow: Few-Step Diffusion Training in ~2 GPU-Hours — bdsqlsz · 2026-10-15
- AI film and TV hit the market: Prime Video picks up AI spin-off, pilots screen at Mipcom — lmoroney · 2026-10-15
- Ramayana film taps DNEG's massive VFX and AI-assisted production for global release — CurieuxExplorer · 2026-10-15