Skip to content
かみもと · 本地AI视频工作流·· 2026-10-19AI score62

Complete guide to training a MOSS-Image character LoRA with ai-toolkit on 12GB VRAM

【VRAM 12GB】画像生成AI「MOSS-Image」でキャラLoRAを作成する完全ガイド(ai-toolkit)

AI briefing

The author uses MOSS-Image-Base with the Qwen3.5-9B text encoder and FLUX.2 VAE to train character LoRAs on a local RTX 4070 12GB setup using 30 animated images and 30 live-action photos respectively, and provides complete ai-toolkit steps and PowerShell execution scripts.

Open original note.com