🔷 Ken and Dakota - Dual Character LoRA — Z-Image

This is a dual-character LoRA trained specifically to test character separation and identity bleed in multi-subject generations.

The two subjects were deliberately chosen to be extreme visual opposites so that any identity crossover becomes immediately obvious:

• Character A: Goth woman wearing black fishnet and dark, high-contrast clothing

• Character B: Handsome man in a clean, tailored business suit with formal styling

If facial features, clothing styles, or overall identity begin to mix — you’ll see it instantly.

This makes the LoRA ideal as a diagnostic and validation tool, not just a character model.

🎯 Intended Use

• Testing dual-character prompts

• Detecting identity bleed

• Validating Z-Image workflows

• Benchmarking multi-subject LoRA behavior

• Training-pipeline experimentation

🔞 SFW / NSFW Notes

This LoRA can be used in both SFW and NSFW scenarios, depending entirely on your prompts and base model.

However:

• It was not specifically trained on NSFW content

• Any NSFW output is a result of model + prompt, not training data

🛠️ Want to Create Your Own Dual-Character LoRA?

This model is an example output of my dual-character training workflow.

If you’d like to build your own (with your own characters, styles, or concepts), the full workflow is available on my Patreon:

👉 workflow covers:

• Dataset structure

• Captioning strategy

• Identity isolation techniques

• Training settings

• Common failure cases (and how to avoid them)