MiniMax H3 workflow: 40s Full HD in ~42 minutes on a 16GB VRAM GPU
Far_Patience4517 · reddit · 2026-10-05
- The author built their first complete from-scratch ComfyUI workflow, generating a continuous 40-second Full HD video in 42 minutes on a 16GB VRAM GPU.
- Core approach: Split Sampling does early denoising at low resolution (much faster), then upscales the latent for high-res refinement; each segment gets an MMH3 Ultimate Upscale pass with temporal chunking and tiled diffusion to keep VRAM manageable.
- Features: two sequential H3 segments with native Motion Context for visual/audio continuity; audio preserved throughout; retry mode for segment 2 only; multi-reference character input via the Deno H3 node; visible controls with internals folded.
- Experimental; workflow shared on Google Drive; results vary with GPU, model versions, LoRAs and duration.
More from Multimodal
- Creator releases radio edit remastered with Suno and Perplexity Computer — inductionheads · 2026-10-05
- Crowd-sourced Westminster photos dropped at exact spots using Google 3D Tiles, no scanning — MonaJalal_ · 2026-10-05
- HeyGen launches Video built on MiniMax H3, generating 720p 15-second clips — HeyGen · 2026-10-05
- Opus vs Astra 'jelly world' islands: better water costs $15.54 vs $10.51 — i_dg23 · 2026-10-05
- Image model A/B tests are back, hinting at a new version in hidden testing — flowersslop · 2026-10-05
- ComfyUI help wanted: MiniMax H3 ref2video with checkpoint and LoRA loaders — aanabobby31 · 2026-10-05