LaunchVisual AIOctober 18, 2026

Fizgig 8.6 improves MOSS-Image LoRA likeness

Read original source →github.com

Fizgig 8.6 now trains block 22 for LoRAs, which was previously skipped as AI Toolkit does; side-by-side runs showed better likeness from the first epochs and more stable results epoch-to-epoch. MOSS-Image support in Fizgig covers LoRAs, LoKRs, sliders, full fine-tuning and multi-subject LoRAs, with LoRAs loading straight into ComfyUI.

2 sources

More stories today

Open the live feed