MiniMax H3 Video on 8GB VRAM: Full ComfyUI Workflow Open-Sourced
Ecstatic-Use-1353 · reddit · 2026-09-04
A developer published an MIT-licensed ComfyUI workflow for running MiniMax H3 FL2V video generation on low-VRAM GPUs:
- Hardware: RTX 4060 Laptop, 8GB VRAM + 16GB RAM
- Key optimizations: W4A8 quantized model, Qwen3-VL 4B INT8 as a smaller encoder replacement, INT8 ConvRot VAE, ClipProj 4B
- Recommended config: 480x864, 158 frames (6.6s at 24FPS), 20-step Euler; 544x960 takes 10-11 min
- Gotchas: 25 steps is unstable on 8GB — stay at 20; higher resolutions work but cost rises sharply
Repo includes install guide, performance notes and troubleshooting.
More from Multimodal
- Artificial Analysis TTS leaderboard: Voxtral 3 leads speed at 112.3 chars/s, Eleven v4 tops quality — ArtificialAnlys · 2026-10-21
- Mistral's Voxtral 3 speech models land #3 on STT leaderboard at $13/1M chars — ArtificialAnlys · 2026-10-21
- A step-by-step system for pumping out hundreds of AI video ads without writing prompts — tyler_agg · 2026-10-21
- H3 keeps racking up distillation LoRAs: PTD lands right after SMD — linoy_tsaban · 2026-10-21
- Reddit's 'The Supply Boat': AI video so lifelike viewers can't tell — Kooky-Parfait-5495 · 2026-10-21
- How I built an end-to-end AI face swap video workflow, from shot planning to frame alignment — Few-Profession421 · 2026-10-21