← Back to Videos

LightVAE Cuts MiniMax H3 Decode Time

Loading video

LightVAE brings pruned lightweight VAE components and an optional INT8 decoder to MiniMax H3 with ComfyUI support; measured decode time drops from 13.28 s on the original INT8 path to 5.71 s with LightVAE Lite.

MiniMax H3 LightVAE Video VAE INT8 Quantization ComfyUI Inference Optimization Video Generation

Category: video-generation

Author: @ITOKtw

Date: 2026-10-15T00:00:00

Duration: 10.125s

Source: https://x.com/ITOKtw/status/2110738842563248480

Reference: https://huggingface.co/lightx2v/LightVAE

OmniHuman 2.0: Audio-Driven Avatar Video at Five Minutes

@BytePlusGlobal · 22 views · 2026-10-15

OmniHuman 2.0BytePlusAudio-Driven Avatars

LongLive: Real-Time Interactive Long Video Generation

LongLive: Real-Time Interactive Long Video Generation

@bkdgiffug · 198 views · 2026-10-11

Video GenerationInteractive VideoAutoregressive

Motion2MixamoRig Retargets Video Motion onto a Mixamo Character

Motion2MixamoRig Retargets Video Motion onto a Mixamo Character

@renmu2017 · 149 views · 2026-10-12

Motion CaptureMixamoBlender

Veda Sparse Attention Speeds Up Video Character Replacement in ComfyUI

Veda Sparse Attention Speeds Up Video Character Replacement in ComfyUI

@ai_hakase_ · 130 views · 2026-10-13

Video GenerationSparse AttentionComfyUI

Seamless 30-second Kung Fu Panda from two MiniMax H3 clips: ComfyUI-H3-Motion-Context chains clips via latent slicing and audio continuation

Seamless 30-second Kung Fu Panda from two MiniMax H3 clips: ComfyUI-H3-Motion-Context chains clips via latent slicing and audio continuation

@eternityspring · 674 views · 2026-10-02

MiniMax H3ComfyUIVideo Generation

OmniChar: One Character Across Every Model, Carried in a Portable .char File

OmniChar: One Character Across Every Model, Carried in a Portable .char File

@aiaicreate · 909 views · 2026-09-28

OmniCharCharacter ConsistencyComfyUI

As an Amazon Associate, we earn from qualifying purchases.