Loading...
Practical, tested guidance on building character datasets that train well: how many images you need, how to caption them, how to stop identity drift, and how to pick between Flux and SDXL.
Written by the LoRA Dataset team. Subscribe via RSS.
Character LoRAs need 15–30 diverse images; styles need 50–200. Learn the ranges by LoRA type, the steps math, and the signs you have too few or too many.
Turn one reference image into a 20+ image, captioned LoRA dataset: reference quality rules, a diversity grid, a curation checklist, and starting settings.
How to write LoRA training captions: choosing a trigger word, the caption-what-changes rule, and caption formats for Flux, SDXL, and SD1.5 with examples.
Why a character LoRA gives a different face on every seed, how to diagnose it with a four-framing grid test, and the dataset and training fixes that work.
Flux.1 vs SDXL for character LoRAs: architecture, VRAM needs, dataset and caption differences, training settings, output quality, and licensing compared.
Upload one reference image and get 20+ identity-consistent, captioned images ready for Flux, SDXL, or SD1.5 training. Free with your own Gemini API key.
Create a dataset