wowsuch ai
Live

Veda-SparseModel

MiniMax-H3 T2VA Veda Predictor 8 NFE 600-Step Preview

This preview checkpoint is a tile-score predictor for MiniMax-H3 text-to-audio-video generation at eight denoising steps. It selects attention tiles for block-sparse attention at a 10% keep ratio, with twelve geometry plans packed into the weights file.

What it does

Use it with MiniMax-H3 to speed up video generation while retaining visual and audio quality comparable to full attention.

Availability
Online demo, ComfyUI custom node, and downloadable checkpoint
Pricing model
Not established
Deployment
Local generation using Miowtion or ComfyUI; GPU requirements apply
Privacy
Not established
Licence
Inherits the MiniMax H3 Community License
Capability
video generation
Modality
videoaudiomultimodal
Access
webopen weights
Deployment
on-premises
Pricing
unknown
Target user
developer

Evidence

4 claims
  • Veda predicts the most important 10% of attention and skips the rest. [1]
    attention and skip the rest.
    Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights.
  • The announcement reports up to 6.8× attention speedup and 3.1× end-to-end speedup. [1]
    - **Faster inference.** Up to 6.8× attention speedup and 3.1× end-to-end
    Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights.
  • The checkpoint is available to download from Hugging Face, and an online demo is provided. [1]
  • The checkpoint is described as applying to T2VA, FL2VA, R2VA, and arbitrary numbers of diffusion steps. [1]
    This checkpoint applies to **all generation modalities** supported by MiniMax-H3, including **T2VA, FL2VA, and R2VA**, as well as to **arbitrary numbers of diffusion steps**.
    Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights.

Sources

1
1
Hugging FaceOwner-controlled
https://huggingface.co/Veda-Sparse/Minimax-H3-T2VA-Veda-8NFE-600Step-Preview Retrieved 2026-10-28 00:02 UTC
Verified

Versions

6
VersionEffectiveChangeChanged fieldsState
62026-10-26 01:12 UTCevidence updatescoreCurrent
52026-10-05 11:45 UTCmaterial changemodality, access, availability, deploymentCurrent
42026-10-04 12:41 UTCmaterial changeavailability, deploymentCurrent
32026-10-01 15:26 UTCmaterial changecapability, modality, availability, deploymentCurrent
22026-09-27 18:59 UTCmaterial changemodality, availability, deploymentCurrent
12026-09-26 19:59 UTCpublicationNoneCurrent

In briefs

2

Open in the catalog

Original catalog content is licensed under CC BY 4.0, attributed to wowsuchai. Source quotations and linked content keep their original rights.