Not loaded
Download
1 variant available
fp16 SafeTensor
ohwxz_person_c1_krea2_turbo.safetensors
Half precision, best balance • 223.81 MB
Verified: 3 days ago

18
18
# ohwx_person (Krea 2 Turbo)
Character LoRA | Base Model: krea-community/Krea-2-Turbo
> Trigger Word: ohwx_person
This is a LoRA for generating a specific AI-generated female character. The model was trained on a dataset of images of a single fictional person and allows you to reproduce her appearance across different scenes, angles, and styles.
This is an AI character. She is not a real person. All training images were AI-generated, and the model does not depict, imitate, or represent any real individual. Any resemblance to actual persons is purely coincidental.
---
## 🖼️ Sample Generation
| Parameter | Value |
| :--- | :--- |
| Prompt | ohwx_person, photo of a young woman, natural face, long dark hair, soft smile, white shirt, standing in a bright room, realistic smartphone photo, natural lighting, high detail, sharp focus |
| Resolution | 768 x 1024 |
| Seed | 1155099827 |
---
## ⚙️ Recommended Settings
Krea 2 Turbo works best with its default sampler settings. For this LoRA, a strength of 0.8–1.0 gives the best balance between likeness and flexibility.
- Steps: 8
- CFG / Guidance: 1.0 (or 0.0 if your interface uses guidance scale)
- Sampler: Euler
- Scheduler: Simple
- LoRA Strength: 0.8–1.0
---
## 🛠️ Training Parameters
| Parameter | Value |
| :--- | :--- |
| Base Model | krea-community/Krea-2-Turbo |
| Trigger Word | ohwx_person |
| Learning Rate | 0.0001 |
| Optimizer | AdamW |
| Network Rank / Alpha | 32 / 32 |
| Dataset Resolution | 1024 x 1024 (bucketing enabled) |
| Bucket Steps | 64 |
| Min / Max Bucket Resolution | 128 / 8192 |
---
## 📝 How to Use
1. Add the trigger. Always start your prompt with ohwx_person to activate the LoRA.
2. Describe the scene. Krea 2 responds best to natural, connected language — describe the scene in full sentences rather than a tag list.
3. Tune the strength. If the likeness is too strong and the character starts to "stick" to one look, lower the LoRA weight to 0.7–0.8. If she doesn't appear clearly enough, raise it to 1.0.
---
## ⚠️ Disclaimer
This model generates an AI character. She is not a real person. All training data was AI-generated, and this LoRA does not depict or impersonate any real individual. Any resemblance to actual persons, living or dead, is purely coincidental. Not intended for use in creating misleading, defamatory, or non-consensual content.
---
## 🙏 Credits
This model received technical support from ModelScope.

