kohya

Bring a folder of images, leave with a trained LoRA. Agent skills (SKILL.md) for quality-gated LoRA training: dataset doctor + safe fixer + one-confirmation training orchestration for SD-Trainer / lora-scripts-next. Anima-first, SD1.5/SDXL/Flux.

Turn a character, style, or concept into a ready-to-train LoRA dataset - consistent multi-angle shots, smart captions (prose / Danbooru / e621 / dedicated taggers), and trainer-ready configs for Flux, SDXL, Krea, Pony & more. Local or cloud, per stage. No training launched, no telemetry.

Published: 2026-10-18

Automate LoRA model training with a guided pipeline for curation, tagging, and validation compatible with Claude Code and other agentic workflows.

Automated Stable Diffusion LoRA pipeline with dataset extraction, training, generation, and evaluation.

LoRA training and sampler ablation study for Stable Diffusion — IEEE 2024

本地 Anima LoRA 训练 WebUI：单图生成素材、一键打标、实时损失曲线、暂停续训，8GB 显存可跑

Local web tool for building and captioning image datasets for LoRA / fine-tuning: LLM query expansion, vision-LLM or WD14 captioning, deduplication. FastAPI + WebSocket; works with Ollama, LM Studio or any OpenAI-compatible API.

只读审计你的 LoRA 训练数据集：caption 遗漏 / 重复 / 触发词漂移 / 扩展名与实际格式不符。零依赖，出一个能双击打开的 HTML。

kohya 风格训练集的数据集标注器：编辑 caption、tag 过滤、WD14 批量打标、生成 dataset.toml · Dataset tagger for kohya-style training sets: edit captions, filter tags, batch WD14 tagging, dataset.toml · kohya 形式の学習セット向けデータセットタガー：キャプション編集、タグ絞り込み、WD14 一括タグ付け、dataset.toml 生成

Minimal UI to train LoRAs with kohya_ss — 6 GB card presets, SD 1.5 & SDXL/Pony, WD14/BLIP auto captions, time estimates, auto copy to Forge. (FR UI, EN summary)