V2:
( all samples upscaled by x1.5 with ciax200K)
Instead of tags, V2 was trained exclusively using natural language captions. The captions were written to capture as much visible skin detail as possible, allowing the LoRA to better understand realistic facial and body features.
V2 focuses on producing a more realistic appearance, especially when prompted for:
Natural skin texture
Shiny or oiled skin
Skin imperfections (only when explicitly prompted)
The dataset also includes a limited number of high-resolution, high-quality backgrounds, although environmental rendering is not the primary focus of this LoRA.
For the best results, prompting using natural language is highly encouraged.
This project aim to explores the limits of realistic rendering with Anima. Results may vary depending on the checkpoint and prompt.
(i don't use ADetailer or any face enhancer)
trigger word: @lifi_RL
Recommended strength: 1.0
V1:
it's aim is to provide more warmer color and lighting on realistic models. On anime-oriented checkpoints such as Janima it tends to push the output toward a semi-realistic aesthetic.
The result highly depend on which checkpoint you use.
Not tested alongside other style Lora so compatibility and results may vary.
trigger word: lifi_rla
Recommended strength: 0.8/1.0
Description
FAQ
Comments (2)
@Lifi_Haf Can You provide some additional information like:
1. what is Your dataset (can You share it?)
2. what did You use for captioning
3. can You share compare with base model used (I seen that base model produce realistic output on its own so what this Lora changes)?
1/
The dataset is AI-generated images enhanced with klein 9B for better face and skin details. Background are copyrighted photography.
2/
Joycaption
3/
The goal of this Lora is to improve skin rendering and push Anima toward a more photorealistic overall appearance (as always, depend on which checkpoint is used) with a slight bias toward curvy figures.