fasth3
Datasets
All datasets matching “fasth3”fasth3-live
FastH3 Live
A continuous, unattended video+audio stream generated by FastH3 (the 4-step DMD2
distillation of MiniMax-H3) on a single consumer GPU, plus everything needed to get
there: the converted weights, a 721-scene prompt library, and the streaming server.
Point VLC at a local URL and it plays without stopping — new clips are generated while
the previous ones play.
Built and measured on one RTX 5090 (32 GB, Windows 11) with ComfyUI.
v1.2.0 — 22.1 fps, up from 18.9, 400 more… See the full description on the dataset page: https://huggingface.co/datasets/jacokon/fasth3-live.fasth3-live-media
FastH3 Live — media
Images for jacokon/fasth3-live,
which is the actual release: converted MiniMax H3 weights, a 321-scene prompt
library, a ComfyUI writer node, and the per-node profiler behind the numbers.
They live here rather than there because that repository is gated on the MiniMax
H3 Community License's territory restriction, and a gate applies to every file in
a repo — so an image linked from it returns 401 to anyone not signed in. These
are original work and need no… See the full description on the dataset page: https://huggingface.co/datasets/jacokon/fasth3-live-media.Minimax_H3_Parasyte_fasth3_6step_testPurpose of this Minimax H3 Turbo lora test is to find the best quality settings for low resolution generation for my old trusted RTX 3060 12Gb. All clips were created with 416x768 resolution.
Workflow used was modified PlagueKind's V7 with Mikey's Wildcard Processor, LLM enhancement node (Minimax-H3-Prompt-Rewriter) and Qwen3VL 4b used as an LLM and text encoder via ClipProj node.
Checkpoint and other settings in the workflow used are by default, or near what Plaguekind has recommended. If… See the full description on the dataset page: https://huggingface.co/datasets/Turpo/Minimax_H3_Parasyte_fasth3_6step_test.fasth3-pr-assets