Charithra Chandran — Krea 2 LoRA
A LoRA trained to reproduce the appearance and facial features of Charithra Chandran with consistent identity across different poses, expressions, outfits, angles, and compositions.
Designed for Krea 2 and trained on a diverse image dataset to capture recognizable facial details, hairstyle, proportions, and overall appearance.
Trigger word
Add chardran as the first token of your prompt:
chardran, a woman with shoulder-length wavy brown hair, light brown eyes, and visible freckles, ...
The trigger must be spelled exactly chardran (lowercase) to activate the LoRA.
Recommended settings
| Setting | Value |
|---|---|
| Base model | Krea 2 Turbo (for inference) |
| Training base | Krea 2 Raw |
| LoRA strength | 1.0 |
| Steps | 8 |
| Guidance scale | 1 (CFG off) |
| mu | 1.15 |
| Resolution | 1024x1024 |
Example generation
Training details
| Parameter | Value |
|---|---|
| Base model | Krea 2 Raw (raw.safetensors) |
| Architecture | Krea-2/lora (networks.lora_krea2) |
| Network dim / alpha | 32 / 32 |
| Training images | 310 |
| Epochs / steps | 5 / 390 |
| Batch size | 1 x grad accum 4 (effective 4) |
| Optimizer | AdamW8bit |
| Learning rate | 1e-4 (constant, warmup 0) |
| Timestep sampling | krea2_shift |
| Weighting scheme | none |
| Resolution | 1024x1024 (bucketed) |
| Mixed precision | bf16 (fp8 base) |
| Seed | 42 |
| Text encoder | Qwen3-VL-4B (frozen) |
| VAE | Qwen-Image VAE |
File
chardran_krea2.safetensors — single-file LoRA, load directly in ComfyUI or with the Krea 2 musubi-tuner inference script.
Model tree for spyrostheboss/Charithra_Chandran_Krea_2_LoRA
Base model
krea/Krea-2-Raw