The LoRA model was built on 60 images taken from publicly available sources on the Internet. LoRA was trained on the FLUX Dev model Improved image quality, works well with other FLUX models.

CFG scale: 2-3.5, experiment with values.

Steps: 14-30, experiment with values.

Weight: 0.4-1.5.

Visible effects will begin to appear at values ​​of 0.4+.

Works well in the range of 0.6-1.2; depending on your preference, experiment with weights up to 1.6. Experiment with values.

Depending on the model, the LoRA weight can be significantly increased to 1.6.

High-quality prompts reduce the negative effect.

Try writing high-quality prompts using different chats: GPT, copilot, grok, gemini, claude, chatgpt, perplexity, deepseek, and so on...