Luna is back — now on Krea2.
The first character I ever released is back, this time trained on Krea2. Same face, same vibe — new base model, new texture.
Luna predates the ShinobiSat Cast (April 2026), and with this release she officially joins it. First character in the series to cross from Z-Image to Krea2.
Specs
Trigger word:
K_LuN4_KSCBase model: Krea2
Recommended weight: 1.0 — but this one behaves a bit differently, see below
Working range: 0.6 → 1.0
Notes on weight
Luna is a distinctive character on her own — she doesn't need to be pushed. A few practical notes:
1.0 and 0.8 are both very faithful — they trade off across scenes rather than one being clearly better. Try both, pick per shot.
0.6 is worth exploring: Luna stays recognizable but reads softer — less made-up, hair less heavy. A "no-makeup Luna" that still holds her identity.
0.4 is edge case — some scenes hold, most lose her.
I wouldn't push above 1.0 — she's marked enough on her own.
Notes
Stress-tested across a wide range of prompts and contexts — character holds, flexibility is there.
All preview images here were generated with Luna V2 alone, no style LoRA stacked. What you see is the LoRA on its own.
Pairs naturally with the ShinobiSat Krea2 Workflow — same pipeline, tested together (link below).
If you cook something nice with her, drop it in the gallery.
Luna V2 is the eighth drop of the ShinobiSat Cast — and the first time an older character crosses over to Krea2 in this series.
— ShinobiSat
Luna — Character LoRA for Z-Image Turbo
Luna is a photorealistic character LoRA trained on Z-Image Turbo. She has short dark wavy hair with bangs, expressive brown eyes, and a naturally intense look that adapts well across editorial, lifestyle, and more creative contexts.
Trained and tested for both SFW and NSFW use. For NSFW content, the standard Z-Image Turbo limits apply — though you may notice some improvements in chest definition.
Trigger Word
ZLuna
Start your prompt with the trigger word, then describe the scene. No need to describe Luna's appearance — the LoRA handles that.
Base Model
Primarily developed and tested on the Z-Image Turbo base model (link below), where she performs best. Other Z-Image Turbo variants have been tested as well — most work fine, a few may need minor adjustments.
LoRA Weight
Recommended range: 0.8 – 1.0
WeightResultBelow 0.6Character identity lost0.6 – 0.8Luna recognizable but fading0.8 – 1.0Sweet spot — full character fidelity1.1 – 1.4Works well, some features more pronounced1.5+Over-saturation, quality degradation
LoRA Compatibility
Tested with several LoRAs — primarily the ones I use most frequently alongside Z-Image Turbo. Some combine seamlessly, others need weight tuning. Multi-LoRA setups always benefit from experimentation.
Known Limitations
Occasional anatomy artifacts (extra limb) may appear during high-volume generation — seems random and not specifically tied to the LoRA. In extensive testing (~200–300 generations), fewer than 5–6 showed any issues.
This is my first character LoRA. Feedback and generations are always welcome — drop them in the comments or tag me!
Description
Specs
Trigger word:
K_LuN4_KSCBase model: Krea2
Recommended weight: 1.0 — but this one behaves a bit differently, see below
Working range: 0.6 → 1.0
Notes on weight
Luna is a distinctive character on her own — she doesn't need to be pushed. A few practical notes:
1.0 and 0.8 are both very faithful — they trade off across scenes rather than one being clearly better. Try both, pick per shot.
0.6 is worth exploring: Luna stays recognizable but reads softer — less made-up, hair less heavy. A "no-makeup Luna" that still holds her identity.
0.4 is edge case — some scenes hold, most lose her.
I wouldn't push above 1.0 — she's marked enough on her own.