Trained using 768x1152 resolution with SD3.5L base model.

TensorArt hosts this Lora in addition to free onsite generation with SD3.5 Large (300 SD3.5 Large images per day for new users , then 100 images per day) : https://tensor.art/models/791213304967242613?source_id=njq1pFzjlEOwpPEpaXny-xcu

//-----//

Training method

You can find the training images + captions here: https://huggingface.co/datasets/codeShare/lora-training-data .

A total of 93 images were used, 10 epochs over 9300 steps.

The prompts were created in JoyCaption:

HF spaces : Joy Caption Pre Alpha - a Hugging Face Space by fancyfeast

Joycaption Notebook : https://colab.research.google.com/github/camenduru/joy-caption-jupyter/blob/main/joy_caption_jupyter.ipynb

and the trigger word 'nihei' was inserted into the resulting prompts at random places.

//----//

How to use:

LoRa looks great in monochrome manga style , but will struggle for other formats.

Themes in LoRA training image prompts: 'monochrome digital art illustration black-and-white drawing manga and anime styles, slender androgynous science-fiction form cybernetic humanoid surreal minimalist geometric woman female creature blades red futuristic tubes metallic organic mutated insect armor high-tech nightmarish helmet gun mechanical appearance urgency robotic tension fitting long hair contrasting gradient sharp weapon post-apocalyptic'