Qwen reference to dataset

Need a dataset to train a character LoRA? This workflow takes one reference image and creates a batch of 12 varied images of the same character, with different outfits, poses and lenses, then adds realistic skin detail.

Great for:

• Training data for an AI influencer LoRA
• Consistent character variations
• Quick lookbooks for one character

How it works

1. Load one reference image.
2. Run it. The 12 prompts are already set up.
3. Review the batch and use the best images for training.

Fast or best quality?

With the Lightning LoRA, Qwen needs only 8 steps at CFG 1, so images come out much faster. Without it, use about 25 steps and a CFG above 1: generation takes longer, but quality is noticeably better.

Make it yours with LoRAs

This workflow supports LoRAs. Grab character, style, anime or cartoon LoRAs from CivitAI or Hugging Face, drop them into the LoRA loader, and keep your AI character looking the same in every shot.