CoolFace
Datasetpublic

supermind/Minimax-H3-sampler_scheduler-tests

Test Configuration Parameter Value Model Minimax H3 FL2VA LoRA Turbo LoRA (LightX2V, converted by Kijai) Steps 6 LoRA Strength 0.75 Optimizations Sage Attention Patch + Mem Eff Sage Attention Patch Tested Combinations The following sampler + scheduler combinations were evaluated: Sampler Schedulers dpmpp_2s_ancestral normal er_sde beta, sgm_uniform, simple euler_ancestral beta, beta57, simple euler_ancestral_cfg_pp beta… See the full description on the dataset page: https://huggingface.co/datasets/supermind/Minimax-H3-sampler_scheduler-tests.

sourceHugging Facemitupdated 2mo agoView on Hugging Face
0likes127downloads
Dataset Card

Overview

This repository contains videos generated with the Minimax H3 FL2VA, with various combinations of samplers and schedulers.

The goal is to identify configurations that produce the most stable, artifact-free, and visually consistent videos.

Test Configuration

ParameterValue
ModelMinimax H3 FL2VA
LoRATurbo LoRA (LightX2V, converted by Kijai)
Steps6
LoRA Strength0.75
OptimizationsSage Attention Patch + Mem Eff Sage Attention Patch

Tested Combinations

The following sampler + scheduler combinations were evaluated:

SamplerSchedulers
dpmpp2sancestralnormal
er_sdebeta, sgm_uniform, simple
euler_ancestralbeta, beta57, simple
eulerancestralcfg_ppbeta, normal, sgm_uniform
eulerbeta, beta57, normal, sgm_uniform, simple
heunpp2beta, beta57, normal, sgm_uniform, simple
jkass_fastbeta, beta57, normal, sgm_uniform
lcmbeta, beta57, normal, sgm_uniform, simple
seeds_2beta, normal, sgm_uniform, simple
res_multistepbeta, beta57, bongtangent, normal, sgmuniform, simple

License

No license

Acknowledgments

  • —LightX2V — Turbo LoRA
  • —Kijai — converted LoRA
  • —Minimax — H3 FL2VA model
  • —jeankassio - jkass fast sampler