First of all, I would like to thank @Ikarirh (<a target="_blank" rel="ugc" href="https://civarchive.com/user/ikarirh">https://civarchive.com/user/ikarirh</a>) for allowing me to use their LoRA as a direct source and inspiration for this LoRA on Krea2.Secondly, I'd like to apologize for the low quality of the example images. I ran out of credits on RunPod, so I had to rush the process. Still, I think the final result turned out well. Mia V1 is a character/identity LoRA for Krea 2 Turbo. It was trained to preserve the identity associated with the trigger token while remaining flexible across wardrobe, location, composition and editorial style.Trigger word: <code>mia_ikari</code> Recommended LoRA weight: <code>0.8</code> Recommended range: <code>0.65–1.0</code> Base model for use: Krea 2 Turbo File: <code>MiaV1_krea2_lora.safetensors</code>This LoRA was trained on Krea 2 Raw and is intended to be used with Krea 2 Turbo. It is not compatible with SDXL, FLUX, Qwen Image or other model families.Example prompttextmia_ikari, editorial portrait of a woman in a cream linen blazer, standing in a modern art gallery, soft window light, medium shot, natural skin texture, 85mm photography.Tips- Start at weight 0.8 and generate a few different seeds.- Use 0.65–0.75 for a more flexible result; use 0.9–1.0 for stronger identity.- Include the trigger only once. Then describe outfit, place, lighting and camera normally.- For portraits, begin at 1024×1024, 8 steps and CFG 1.0 in ComfyUI's generic KSampler.- If clothing or a prior scene appears too strongly, lower the weight and explicitly state the desired wardrobe/background.Training notes- 116 curated image/caption pairs- 1024×1024 bucketed training, without upscaling- Rank 32 / alpha 32- AdamW 8-bit, learning rate 1e-4- 2500 steps, batch size 1, seed 42- Trained with Musubi Tuner on an NVIDIA A100 80 GBFiles and integrity| File | Role | SHA-256 || <code>MiaV1_krea2_lora.safetensors</code> | Final 2500-step release | <code>76d2c86351cc49de204782a8c6807707973282e1e6632353f01985466e9b954c</code> || <code>MiaV1_krea2_lora-step00001500.safetensors</code> | Optional comparison checkpoint | <code>956002f20534936d7156fe1d94608c46a178dcca892e0d81fb371010afe32c74</code> || <code>MiaV1_krea2_lora-step00002000.safetensors</code> | Optional comparison checkpoint | <code>e4e9c1feeedeaf22c10ab4ef0cdefc17a5465efe89ff7273236db2ecfaa76b69</code> || <code>MiaV1_krea2_lora-step00002500.safetensors</code> | Optional final-step checkpoint | <code>18b9c7eb5e315e9ec93b84a6bc103b8657ac147bf0db39558461866999c7575c</code> |Responsible useOnly use or redistribute this LoRA where you have the required rights and consent for the reference images and represented identity. Do not use it for impersonation, deception, harassment, non-consensual intimate imagery, defamation or privacy violations. Use is also subject to the Krea 2 Community License and acceptable-use terms.

Mia V1 — Krea 2 Character LoRA

Mia V1 is a character/identity LoRA for Krea 2 Turbo. It was trained to preserve the identity associated with the trigger token while remaining flexible across wardrobe, location, composition and editorial style.Trigger word: <code>mia_ikari</code> Recommended LoRA weight: <code>0.8</code> Recommended range: <code>0.65–1.0</code> Base model for use: Krea 2 Turbo File: <code>MiaV1_krea2_lora.safetensors</code>

Files with SHA256: 76d2c86351cc49de204782a8c6807707973282e1e6632353f01985466e9b954c