Maestro 3.2's hidden FastH3 toggle: 8-step text-to-video runs in 11 minutes
cocktailpeanut · x · 2026-10-23
Pinokio creator cocktailpeanut notes that Maestro 3.2's experimental FastH3 model must be enabled manually: go to "Enable more models" and check "FastH3 V2 8-Step -- Text to Video (Experimental)".
FastH3 V2 is an experimental eight-step text-to-video option with generated audio; one user reported 11m13s per generation on an A4500. The release also adds OmniRef (image/video/audio references with BF16, INT8, FP8 checkpoints) and an NVFP4 quantized model.
Related event: FastH3 V2 Generates 15-Second Video Locally in 11 Minutes on One GPU(3 posts)→
More from Multimodal
- Tencent releases WeVisOne-VL: one VLM for detection, segmentation, OCR and matting — _akhaliq · 2026-10-23
- Tencent releases Hy-Nomas-Bench: 280 cases testing video world model controllability — _akhaliq · 2026-10-23
- Kling 4.0 hits Runware API day one: 30s 1080p video with native audio from $0.133/s — aziz4ai · 2026-10-23
- Free realistic portrait model Jib Mix Moss Realistic ships with 8-step merged LoRA — jib_reddit · 2026-10-23
- huashu-design 3.0 released: open-source design skill hits 25k stars after three iterations — AlchainHust · 2026-10-23
- MOSS-Image 5B finetune hits ~90% artist-style accuracy with just 3,685 images on one MI300X — Turbulent-Bass-649 · 2026-10-23