Search results

Huge and sustained loss after resume wan 2.2 i2v training lora

Status: Open.

#667 In kohya-ss/musubi-tuner;

· bazanga912 opened on Oct 21, 2025

Speed decrease over time

Status: Open.

#665 In kohya-ss/musubi-tuner;

· GreenLandisaLie opened on Oct 20, 2025

Wan 2.2 is trained in FP16 ? FP16 is way weaker to BF16 no one tried bf16 base models?

Status: Open.

#664 In kohya-ss/musubi-tuner;

· FurkanGozukara opened on Oct 20, 2025

Qwen Image Fine tuning metadata base resolution is inaccurate

Status: Open.

#663 In kohya-ss/musubi-tuner;

· FurkanGozukara opened on Oct 19, 2025

Help - Training WAN 2.1 or 2.2 Lora Models

Status: Open.

#662 In kohya-ss/musubi-tuner;

· holycowdude opened on Oct 19, 2025

Hi, I trained a low-noise character LoRa on a Wan2.2 T2V low model. It works great in the images generated during procedural sampling, but it doesn't work when used in ComfyUI.

Status: Open.

#661 In kohya-ss/musubi-tuner;

· icefox21 opened on Oct 19, 2025

ZeroDivisionError: integer division or modulo by zero

Status: Open.

#660 In kohya-ss/musubi-tuner;

· FurkanGozukara opened on Oct 19, 2025

VRAM Usage Doubles When Using DDP in Qwen-Image Training

Status: Open.

#658 In kohya-ss/musubi-tuner;

· kaiyama12345679 opened on Oct 18, 2025

src/musubi_tuner/qwen_image_cache_latents.py is not obeying the given cuda index

Status: Open.

#657 In kohya-ss/musubi-tuner;

· FurkanGozukara opened on Oct 17, 2025

Video frame extraction ignores Sample Aspect Ratio (SAR), causing training on distorted frames

Status: Open.

#656 In kohya-ss/musubi-tuner;

· diodiogod opened on Oct 17, 2025

Something so wrong with Loading DiT model from disk logic

Status: Open.

#655 In kohya-ss/musubi-tuner;

· FurkanGozukara opened on Oct 17, 2025

[BUG]Qwen-edit-plus training with sample

Status: Open.

#654 In kohya-ss/musubi-tuner;

· sdbds opened on Oct 17, 2025