The goal is to capture the essence of who Fern is, her anatomy, such that generated content has higher parity with Fern character, including explicit parts. Which means, T2V and I2V will be of higher quality and better fidelity to the character. This model is not aiming at any particular effects, transitions, camera movement, etc.

High quality images were used to help the model to have less noise and achieve higher quality, the goal is to not depend on other LoRAs for high quality, better esthetic or NSFW generation, avoiding LoRA accumulation errors and saving memory.

The model was trained for Wan 2.1 14b parameter and aims at high quality.

1. Dataset of 220 images

2. 1280x720 and 720x1280 resolution

3. Complex prompt created by florence2

4. over 10000 steps over 45 epochs.

In the dataset there's a collection of Ferns vanilla art, NSFW art but also some human dressing as Fern.

Trigger Words are: fern from sousou no frieren, hfernai

Example prompt: fern from sousou no frieren, hfernai, a digital illustration shoot from behind about a character with long purple hair in a suggestive pose, wearing a blue thong, in an indoor setting with a softly lit room and a bed in the background. the image also shows a focus on the character's back and lower body. on the middle of the image, a 1woman, who appears to be in her early twenties, is standing with her back to the viewer, looking over her shoulder with a sultry expression. she has long, straight, purple hair with a hair clip, and is wearing a black thong. her body is slim, and she has a mole on her ass. the lighting is soft and warm, casting gentle shadows on her body. the background is simple, with a bed and a window letting in natural light. the style is anime-inspired, with smooth shading and vibrant colors.