ZImage-Base Turbo LoRA - Experimental

Goal

A Turbo LoRA for ZImage-Base that gets as close as possible to ZImage-Turbo quality.

Wish me luck 😉

Roadmap

v0 (currently here)

• ~2K training prompts

• Mostly female portraits

• Establish baseline quality and validate training setup

v0.5

• ~20K diverse prompts

• Evaluate scaling behavior and improve generalization

v1

• ~200K diverse prompts

• Target significant quality gains and approach ZImage-Turbo performance

  Future

• Scale to 2M prompts if results continue to improve

• Aim to match or surpass ZImage-Turbo quality

FAQ

Will this work with other Loras?

-> Hopefully 🤞. If it does let me know.

Will this work with Zit?

-> Yes but why?