Downloads · 30 days
0
laoda/DasiwaMinimaxH3_dasiwaHybrid8turboV1
DasiwaMinimaxH3_dasiwaHybrid8turboV1 is a text-to-video model from laoda. Use it when you need video from a text prompt. It is set up for comfyui.
A merged MiniMax H3 checkpoint built for 8-step sampling with the ComfyUI DaSiWa node pack (ComfyUI-DaSiWa-Nodes, MiniMaxH3Director).
Downloads · 30 days
0
Access
Public
Updated Sep 9, 2026
Repo size
21 GB
Likes
1
Public
Click a slice to open those files.
.safetensors21 GB · 100%
From the Hugging Face model README
A merged MiniMax H3 checkpoint built for 8-step sampling with the ComfyUI
DaSiWa node pack (ComfyUI-DaSiWa-Nodes, MiniMaxH3Director).
| file | size | notes |
|---|---|---|
DasiwaMinimaxH3_dasiwaHybrid8turboV1.safetensors | 20,967,642,441 B (19.5 GiB) | load with UNETLoader from models/diffusion_models/ |
REF2VA (reference-image to video+audio). FL2VA is wired in the same
workflow but was not exercised in our tests.MiniMaxH3SigmaShift: shift_video = 10frame_rate: 24ref_image_size: match — with 1024×768 reference images the output renders
960×704; portrait output needs portrait references or a canvas override.Optional LoRA: the DaSiWa workflow exposes a LoRA slot that references
MiniMax-H3-Ref2VA-Acc-8Step.safetensors. That file was not present in our
runs (the loader logs LoRA not found) and generation still succeeded — this
checkpoint does not require it.
Tested on an RTX PRO 6000 Blackwell (97 GB) with ComfyUI + sage attention:
overall_soundscape and
non_diegetic_music in the six-section prompt for dialogue delivery and score.DaSiWa MiniMax H3 Director (ComfyUI custom nodes) — see the node pack's README.