Gwangya Pretty Anime is a style LoRA for Krea 2, focused on expressive anime portraits, flowing hair, rich colors, and a balance of soft facial shading and defined anime shadow shapes.
In the current female portrait tests, the LoRA produced narrower eyes, deeper facial shading, fuller hair, and stronger color contrast compared with the base model.
Tested setup
• Base model: krea2_turbo_fp8_scaled.safetensors
• Text encoder: qwen3vl_4b_fp8_scaled.safetensors
• VAE: qwen_image_vae.safetensors
• LoRA strength: 0.8
• Sampler: Euler
• Scheduler: Simple
• Steps: 8
• CFG: 1
• Resolution: 1344 × 752
• No reference image or additional LoRA was used in these comparisons.
Start with a LoRA strength of 0.8 and adjust it to suit your prompt. This is a tested starting point, not a universally optimal setting.
Limitations
Dark outlines and cel-shaded boundaries can remain. Results vary with the prompt and seed. The current comparison covers two seeds of an adult female portrait prompt; broader character and scene consistency has not yet been established.
The comparison images use the same prompt, seed, resolution, and sampling settings within each pair, changing only whether this LoRA is applied.
This is an independently created LoRA adaptation for Krea 2, not an official Krea product. Use and redistribution remain subject to the Krea 2 Community License and its applicable conditions.
Description
Initial release of Gwangya Pretty Anime for Krea 2.
Trigger word: wil1lora
Training steps: 4,000
Training epochs: 33
Tested generation settings:
• Krea 2 Turbo FP8
• LoRA strength: 0.8
• Euler / Simple
• 8 sampling steps
• CFG: 1
• Resolution: 1344 × 752
The current comparison examples use two matching seed pairs, with and without the LoRA. They were generated without the trigger word; trigger-enabled results have not yet been compared.
Dark outlines and cel-shaded shadow boundaries may remain. Results depend on the prompt and seed.





