MiniMax H3 Runs Locally on a RTX 5060 Ti 16GB — Full ComfyUI Setup Shared
Description-Serious · reddit · 2026-11-07
A Reddit user got MiniMax H3 video generation running locally on consumer hardware and shared the full recipe:
- Hardware: RTX 5060 Ti 16GB + 32GB RAM, Windows, ComfyUI Desktop. VRAM hit the full 16/16 GB and RAM peaked around 96% during generation.
- Model stack: minimaxh3fl2vaprunedfp8scaled base model, NVFP4 AWQ text encoder, Turbo 4-step LoRA (0.9) plus a facial realism LoRA (0.9).
- Settings: euler sampler, beta scheduler, 4 steps, 9:16 at 0.4 MP, 8 s at 24 fps, refimagesize max, fixed seed.
The screenshot shows surprisingly solid character video quality. The author is now asking how to upscale output without blowing up the PC — a directly reproducible reference for anyone wanting local video gen on 16GB VRAM.
More from Infra
- Decision models fix a pricing inefficiency: prefill workloads subsidize decode, argues sh_reya — hugobowne · 2026-11-07
- Team weighs multi-GPU cluster vs one fat workstation to escape cloud API rate limits — Decent_Wing_4393 · 2026-11-07
- AMD's Halo Station packs 96 cores, 2TB RAM, 576GB GPU memory — for $100k-$150k — Sea_Emergency_8458 · 2026-11-07
- Satnam Singh: AI Is Dissolving the Hardware/Software Boundary for Good — satnam6502 · 2026-11-07
- Data centers' water use is just ~4% of US golf course irrigation, study argues — kimmonismus · 2026-11-07
- Data centers' water use vastly overstated: just 4% of US golf course irrigation — kimmonismus · 2026-11-07