app download link : https://www.patreon.com/SECourses/posts/secourses-musubi-137551634

SECourses Musubi Trainer V32 — Official 4K App Screenshots

Used in the article : https://www.patreon.com/SECourses/posts/secourses-musubi-137551634

Qwen Image Training tab

Inside the Qwen tab 1/6 — config presets, Accelerate launch and checkpoint/resume management

Inside the Qwen tab 2/6 — dataset preparation with automatic TOML generation

Inside the Qwen tab 3/6 — model settings, Torch Compile and caching

Inside the Qwen tab 4/6 — optimizers, scheduler and LoRA architecture

Inside the Qwen tab 5/6 — training settings and sample generation

Inside the Qwen tab 6/6 — advanced, metadata and Hugging Face upload

Wan 2.1 / 2.2 Models Training tab

FLUX 2 Dev + FLUX Klein 9B / 4B Training tab

Z-Image Base & Turbo Training tab

Ideogram 4 Training tab

Krea 2 Training tab

Krea 2 Model Settings — exclusive INT8 ConvRot quantized training

LTX 2.3 / LTX-2 Video Training tab — new in V32

Qwen2.5-VL Image Captioning tab

Model Quantizer tab — FP8 Scaled / INT8 ConvRot / NVFP4 / MXFP8

LoRA Extractor tab

LoRA Merger tab

LoRA Converter tab

Image Preprocessing tab

Version History tab — 42 releases with accordion changelogs

SECourses Musubi Trainer vs stock Musubi Tuner comparison

Supported models & training modes matrix

Training speed-up benchmark chart

V32.1 Modernized Interface — Per-Tab Section Screenshots (4K)

Wan tab 1/7 — configs, Accelerate and checkpointing

Wan tab 2/7 — video dataset preparation with frame extraction controls

Wan tab 3/7 — built-in complete Wan dataset guide

Wan tab 4/7 — model settings incl. Wan 2.2 dual-model system

Wan tab 5/7 — timestep sampling, loss weighting and CUDA options

Wan tab 6/7 — training, network, optimizer and caching

Wan tab 7/7 — samples, advanced, metadata and Hugging Face

FLUX tab 1/3 — model family, dataset and model settings

FLUX tab 2/3 — Torch Compile, training and sampling

FLUX tab 3/3 — optimizer, LoRA and advanced

Z-Image tab 1/3 — configs, dataset and model settings

Z-Image tab 2/3 — Torch Compile, training and sampling

Z-Image tab 3/3 — optimizer, LoRA and advanced

Ideogram 4 tab 1/3 — configs, dataset and model settings

Ideogram 4 tab 2/3 — flow matching, training and sampling

Ideogram 4 tab 3/3 — LoRA, advanced and Hugging Face

Krea 2 tab 1/3 — configs, dataset and INT8 ConvRot model settings

Krea 2 tab 2/3 — flow matching, training and sampling

Krea 2 tab 3/3 — LoRA, advanced and Hugging Face

LTX 2.3 tab 1/3 — model, quantization and video dataset

LTX 2.3 tab 2/3 — audio, caching, network and training

LTX 2.3 tab 3/3 — logging, metadata and one-click start

Krea 2 INT8 ConvRot measured end-to-end training speed table