Sponsorship: For sponsorship inquiries, please reach out to us on Discord. Your support helps us grow and improve our future work.

Buy Custom Models: Interested in purchasing a custom LoRA model? Contact us we can deliver quickly.

Niji Style FLUX

I’m excited to share my new model, Niji Style FLUX. This model is crafted to generate stunning Niji-style images, known for their vibrant colors and expressive, anime-inspired art.

Model Details:

Type: Niji-style image generation
Trigger Words: None required, but using them may enhance results.
Output: High-detail, vibrant, and expressive Niji-style images that truly capture the essence of anime art.
NSFW Support: No

How I Made This Model:

Following a proven approach, I trained a LoRA model specifically for Niji-style artwork. Afterward, I merged it with a base model to optimize for the unique characteristics of Niji art.

Training Configuration:

GPU: A6000
Dataset Size: 2,000 images
Captioning Tool: GPT-4
Batch Size: 1
Optimizer: AdamW
Learning Rate: 1e-4
Total Training Steps: 25k
Training Duration: 24+ hours

How to Use This Model:

This model is versatile and can be easily integrated into your preferred setup. If you're familiar with my previous models, you'll find this one just as user-friendly.

Quick Parameters:

Sampler: Euler
Scheduler: Simple
Sampling Steps: 20 (If using Schnell, steps should be set to 4)
CFG Scale: 1

Model Versions:

FP8 Model (No CLIP and VAE): 11.08GB – Ideal for users with lower GPU power, offering efficiency while maintaining quality.
FP8 Model with Baked VAE & CLIP: 15.91GB – A balanced option with baked-in components for those who prefer an all-in-one model.

Note:

If you downloaded the 11.08GB model, move it to ComfyUI\models\unet after downloading.
If you downloaded the 15.91GB model, move it to ComfyUI\models\checkpoints after downloading.

Join My Community: Share your feedback, learn, and have fun with us! 😊

Discord

Niji Style - V0.5