I built this because creating prompts for LoRA training datasets is incredibly repetitive, especially when you need prompts for hundreds of images. Most people either reuse the same prompt with small tweaks or write everything manually, which leads to poor variation and weak training results. The model doesn’t learn properly because the dataset lacks structure and diversity. This platform solves that by generating structured prompts across multiple images. You can generate a full dataset with controlled variation. Initially, I was just trying to speed things up. But while building, I realized the real issue wasn’t speed, it was how hard it is to maintain balance between consistency and variation across a dataset. So the product evolved into something more focused on generating prompts that actually work for training, not just for images. The goal is simple: help people build stronger LoRA datasets with less effort and better results.

Ink and Lens

Prompt engineering for LoRA datasets

Prompt engineering for LoRA datasets

Forum Threads

Feedback

Forum Threads

Feedback