YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
JHVKP7 Realism β Janhvi Kapoor | Krea 2 Character LoRA
Photorealistic Character LoRA | Krea 2 RAW | 1,200 Training Steps
JHVKP7 Realism is an independently trained, unofficial character LoRA based on Krea 2 RAW, developed to reproduce the recognizable facial characteristics of Indian actress Janhvi Kapoor in synthetic AI-generated imagery.
The project focuses on facial identity consistency, natural-looking skin textures, realistic facial anatomy, and flexible photographic generation across different environments, outfits, lighting conditions, and camera angles.
The LoRA was trained for 1,200 optimization steps using a curated dataset of 55 character photographs.
β¨ Key Features
- Identity-focused training: Learns distinctive facial characteristics and proportions.
- Photorealistic portraits: Designed for believable skin textures, realistic lighting, and natural facial details.
- Facial expression variety: Intended to support neutral expressions, candid smiles, and expressive portraits.
- Fashion versatility: Suitable for experimenting with traditional Indian outfits, casual clothing, and editorial photography.
- Flexible compositions: Supports close-up portraits, half-body compositions, and full-body photography.
- Krea 2 ecosystem: Designed for Krea 2 RAW with potential compatibility across supported Krea 2 variants.
These are training objectives, not independently benchmarked performance guarantees.
𧬠Technical Details
| Parameter | Information |
|---|---|
| Model name | JHVKP7 Realism |
| Model type | Character / Identity LoRA |
| Character | Janhvi Kapoor |
| Base checkpoint | Krea 2 RAW |
| Training framework | Musubi Tuner |
| Training steps | 1,200 |
| Training images | 55 curated photographs |
| Trigger token | jhvkp7 |
| Model format | .safetensors |
| Task | Text-to-image generation |
| Training objective | Photorealistic identity consistency |
This repository provides LoRA adapter weights. The corresponding Krea 2 base checkpoint is required for inference.
π How to Use
- Download the LoRA
.safetensorsfile. - Load a compatible Krea 2 base checkpoint in your inference workflow.
- Apply the LoRA using the appropriate Krea 2 LoRA loader.
- Include
jhvkp7in your text prompt. - Start with a LoRA strength of 0.8 and adjust as needed.
Suggested LoRA strength: 0.6β1.0
Recommended starting strength: 0.8
Trigger token: jhvkp7
The trigger token was used in the training captions. It is recommended for prompting, but the adapter can also influence generation when loaded without the token.
πΈ Sample Prompts
Natural Portrait
Photorealistic candid portrait of jhvkp7, natural expression, soft window light, realistic skin pores, subtle facial imperfections, detailed eyes, natural skin tones, shallow depth of field, 85mm lens, professional photography.
Traditional Indian Fashion
A realistic editorial photograph of jhvkp7 wearing an elegant Indian saree, standing on a terrace in soft morning sunlight, natural facial details, realistic fabric textures, gentle shadows, authentic colors, professional fashion photography.
Outdoor Candid Photography
Natural candid photograph of jhvkp7 walking outdoors, casual everyday clothing, relaxed expression, sunlight filtering through trees, documentary photography, 35mm lens, realistic skin details, natural body proportions.
Cinematic Portrait
Cinematic photograph of jhvkp7 sitting inside a modern cafΓ©, soft evening light, realistic facial anatomy, warm atmospheric lighting, subtle film grain, shallow depth of field, premium photographic quality.
Studio Fashion Photography
Professional editorial portrait of jhvkp7, neutral studio background, soft diffused lighting, elegant contemporary outfit, realistic skin texture, sharp facial details, balanced exposure, professional fashion photography.
βοΈ Checkpoint Compatibility
Primary base model: Krea 2 RAW
Other checkpoints to test:
- Krea 2 Turbo
- Krea2-SAT-DirtyRealism
- Other architecturally compatible Krea 2 variants
The LoRA was trained on RAW. Performance and compatibility on third-party checkpoints are experimental and should be independently verified.
π¬ Training and Evaluation
The adapter was trained with an identity-focused photographic dataset containing different facial angles, expressions, clothing, and lighting conditions.
Evaluation should consider:
- Facial likeness and consistency
- Natural skin texture
- Prompt adherence
- Expression flexibility
- Performance across multiple camera angles
- Generalization to unseen backgrounds and outfits
β οΈ Known Limitations
- Facial likeness can vary depending on the prompt, checkpoint, and LoRA strength.
- Complex poses and extreme camera angles may introduce identity drift.
- Excessive LoRA strength may reduce prompt flexibility.
- Differences between Krea 2 variants can affect image quality.
- Generated images may include unrealistic details or visual artifacts.
π‘οΈ Responsible Use
This is an unofficial AI character research project and is not affiliated with, authorized by, or endorsed by Janhvi Kapoor.
Generated images must be identified as synthetic AI content and should not be represented as authentic photographs, public statements, or endorsements.
Do not use the model for deceptive impersonation, fraud, harassment, or non-consensual intimate imagery.
Users are responsible for ensuring that their use, distribution, and any commercial activity comply with applicable laws, image rights, training-data permissions, and base-model licensing terms.
π¬ Community Feedback
Feedback and comparisons are welcome.
JHVKP7 Realism β Identity-focused character generation for the Krea 2 ecosystem.


