First-time trainer open-sources full recipe for a Japanese gravure-style Krea 2 LoRA: 9k photos, VLM captions, all params
Zealousideal_Knee_50 · reddit · 2026-10-06
A first-time trainer shares a complete recipe for a Japanese gravure photo-style LoRA on Krea 2: 9,000 adult-subject photos captioned by a VLM with 120–220-word structured descriptions, watermarks cropped, trained with ai-toolkit (rank 32, lr 1e-4, 10k steps, flowmatch). Includes trigger word usage, stacking weights with character LoRAs, natural-language prompting order, known limits, and a linked two-pass refine workflow. 18+ content.
More from Multimodal
- ComfyUI queue stuck? A maintainer's checklist to separate validation, node failures and lost progress — fluxdraw · 2026-10-06
- First try with Seedance 2.5: the model butchered the on-screen text at the end — atomantsmasher · 2026-10-06
- OneCanvas turns a single image into 3D spatial reasoning, hitting new SOTA — Roger_M_Taylor · 2026-10-06
- WM-VLM: world model generates visual intermediate states for spatial reasoning — ZhitingHu · 2026-10-06
- ProAR turns autoregressive video models into goal-oriented prospective reasoners — muhao_chen · 2026-10-06
- Stanford's SPEED diffusion sampler accepted at NeurIPS, ~2X speedup training-free — hsu_byron · 2026-10-06