Sura is an original AI character (fictional, created from scratch, not based on any real person): long straight jet-black hair with a centre parting, two pink hair clips, brown eyes with winged eyeliner, full glossy lips, small hoop earrings, a black choker and fine-line bow tattoos on both forearms.
Trigger word: sura. The trigger carries her face, hair, clips, choker and tattoos, so describe only the scene, the outfit, the pose and the expression. Under strongly coloured light (neon, sunset) add "black hair", otherwise the light can tint her hair.
Versions: one per base model. The anime versions (Pony V6, Illustrious, NoobAI, Anima) were trained on 2D redraws of the same photo set; SDXL, Krea 2 and Z-Image Turbo give the photo Sura; Qwen-Image-Edit 2511 puts her into an existing picture.
Suggested prompts. Pony: score_9, score_8_up, score_7_up, sura, 1girl, solo, [outfit], [place], looking at viewer. Illustrious / NoobAI: masterpiece, best quality, sura, 1girl, solo, [outfit], [place]. SDXL / Krea 2 / Z-Image: sura, a young woman [doing what], [outfit], [place], [light].
Weight 0.8-1.0. Clips, choker and tattoos can be removed by naming them (no hair clips, no choker).
Description
Anime Sura for Anima (anima_base_v10, Qwen3-0.6B text encoder). Trained with sd-scripts (networks.lora_anima) on 32 tagged images, 6 epochs, dim 16 / alpha 16, unet only. Tag prompt: sura, 1girl, black hair, hairclip, black choker, hoop earrings, [scene tags]. Weight 0.8-1.0. Loads in ComfyUI: UNETLoader anima_base_v10 + CLIPLoader qwen_3_06b_base (type cosmos) + qwen_image_vae.
FAQ
Comments (3)
great
How do you train your LoRA? Its anatomy is completely broken. If I use any other character—even without a LoRA—everything turns out perfectly, whereas yours is a textbook example of mutation and deformity. On top of that, the tattoos bleed over onto male characters; I tried generating black-furred demons, but they ended up with bald patches and visible colored tattoos on their dark skin. No amount of tweaking fixes it. Honestly, I couldn't find a single usable shot among all the generations. A prime example of "amazing" LoRA training.
i will fix it, it my first lora, :)