|
--- |
|
license: other |
|
license_name: flux-1-dev-non-commercial-license |
|
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md |
|
language: |
|
- en |
|
tags: |
|
- flux |
|
- diffusers |
|
- lora |
|
- replicate |
|
- anime |
|
- text-to-image |
|
base_model: black-forest-labs/FLUX.1-dev |
|
pipeline_tag: text-to-image |
|
instance_prompt: animexl |
|
widget: |
|
- example_title: Futuristic Samurai |
|
inputs: |
|
prompt: animexl a cyberpunk samurai standing on a neon-lit rooftop, wearing black armor with glowing blue patterns, holding a katana with an electric blade, a bustling futuristic city in the background with flying cars and holographic billboards, cinematic lighting |
|
outputs: |
|
- image: https://example.com/samurai.png |
|
- example_title: Magical Forest Scene |
|
inputs: |
|
prompt: animexl an ethereal elf princess with glowing silver hair, wearing a flowing emerald gown adorned with golden patterns, sitting by a glowing blue lake in an enchanted forest, surrounded by bioluminescent flowers, glowing fireflies, and ancient stone ruins covered in moss |
|
outputs: |
|
- image: https://example.com/forest.png |
|
--- |
|
|
|
# AnimeXL - Fine-Tuned LoRA for Anime Image Generation |
|
|
|
AnimeXL is a LoRA fine-tuned using 200+ anime-style images over 2500 training steps. It specializes in creating vibrant, highly detailed anime-style visuals. The model leverages the **FLUX.1-dev** base model and is compatible with the 🧨 `diffusers` library for seamless integration. |
|
|
|
--- |
|
|
|
## Features |
|
- **Trigger Word:** Use `animexl` in prompts to activate the model. |
|
- **High Compatibility:** Fully integrated with `diffusers` and works with FLUX.1-dev. |
|
- **Specialized Output:** Designed to generate anime-style artwork, perfect for creative projects. |
|
|
|
## How to Use |
|
|
|
### Using 🧨 Diffusers |
|
```python |
|
from diffusers import AutoPipelineForText2Image |
|
import torch |
|
|
|
# Load the base pipeline |
|
pipeline = AutoPipelineForText2Image.from_pretrained( |
|
"black-forest-labs/FLUX.1-dev", |
|
torch_dtype=torch.float16 |
|
).to("cuda") |
|
|
|
# Load the LoRA weights |
|
pipeline.load_lora_weights("Nishit2411/AnimeXL", weight_name="lora.safetensors") |
|
|
|
# Generate an image |
|
prompt = "animexl a futuristic city with anime characters" |
|
image = pipeline(prompt).images[0] |
|
|
|
# Save or display the image |
|
image.save("output.png") |
|
|
|
Trigger Words |
|
Use animexl in prompts to achieve optimal results. Example prompts: |
|
|
|
animexl a magical girl in a serene forest |
|
animexl a samurai standing under a cherry blossom tree at sunset |
|
About Training |
|
AnimeXL was trained on Replicate's LoRA Trainer using the following configuration: |
|
|
|
Base Model: FLUX.1-dev |
|
Training Steps: 2500 |
|
Images: 200+ anime-style images |
|
Resolution: 512x512, 768x768, and 1024x1024 |
|
Batch Size: 1 |
|
Optimizer: AdamW8bit with a learning rate of 0.0004 |
|
Scheduler: Flowmatch |
|
Sampling: Flowmatch with 28 steps, guidance scale of 3.5 |
|
For more details, refer to Replicate's LoRA Trainer Documentation. |
|
|
