Hugging Face's logo

Tech-Meld
/
Life-XL_V1

Text-to-Image
Diffusers
Safetensors
English
StableDiffusionXLPipeline
stable-diffusion
stable-diffusion-xl
lora
realvis
life-fx
Inference Endpoints
Model card Files Community
Life-XL_V1 / README.md
Tech-Meld's picture
Tech-Meld
Update README.md
390e596 verified
preview code
|
raw
history blame contribute delete
4.59 kB
metadata
license: apache-2.0
language:
  - en
library_name: diffusers
tags:
  - stable-diffusion
  - text-to-image
  - stable-diffusion-xl
  - lora
  - realvis
  - life-fx

RealVis XL 4.0 Enhanced with Life FX 1.5

This model is a powerful fusion of the RealVis XL 4.0 base model, renowned for its photorealism, and the Life FX 1.5 LoRA, celebrated for its vibrant colors, textures, and subtle surrealism. This merge unlocks a new level of creativity, allowing you to generate images with stunning realism while infusing them with a unique artistic flair.

image 1 image 2 image 3 image 4 image 5

Model Capabilities:

This model excels at generating images with the following qualities:

Photorealism: Inherit the photorealistic rendering capabilities of RealVis XL 4.0, capturing intricate details, lighting, and shadows.

Enhanced Colors and Textures: Experience the vibrant color palettes and emphasis on texture provided by Life FX 1.5. Expect images with depth and visual interest, pushing beyond standard photorealism.

Subtle Surrealism: The touch of surrealism from Life FX 1.5 introduces an element of the unexpected, adding a layer of intrigue and creative possibility to your generated images.

Strong Compositions: Generate images with well-balanced compositions, clear focal points, and a strong sense of visual storytelling.

Versatile Subject Matter: Tackle a wide range of subjects, from portraits and landscapes to abstract concepts and imaginative scenes.

Usage Instructions:

You can use this merged model using the diffusers library and the peft library (for loading the LoRA).

1. Installation:

pip install diffusers transformers accelerate peft

2. Load the Model:

import torch
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
from peft import PeftModel 

# Model IDs
base_model_id = "stabilityai/stable-diffusion-xl-base-1.0" 
lora_id = "Tech-Meld/Life-XL_V1"
lora_adapter_name = "YOUR_LORA_ADAPTER_NAME"  # Replace with the actual adapter name

# Load the pipeline
pipe = DiffusionPipeline.from_pretrained(
    base_model_id,
    torch_dtype=torch.float16,
    variant="fp16",
    use_safetensors=True,
).to("cuda")

pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)

# Load the LoRA weights
pipe.unet = PeftModel.from_pretrained(
    pipe.unet, 
    lora_id, 
    torch_dtype=torch.float16, 
    adapter_name=lora_adapter_name
)

3. Generate Images:

prompt = "A photorealistic portrait of a woman with vibrant blue hair, standing in a field of sunflowers."
lora_scale = 0.8  # Adjust the LoRA scale (0.0 to 1.0)

image = pipe(
    prompt, 
    num_inference_steps=30, 
    cross_attention_kwargs={"scale": lora_scale},
    generator=torch.manual_seed(0)
).images[0]

image.save("Life-XL_generated.png")

4. Tips and Considerations:

  • Negative Prompt: Using a negative prompt with RealVis XL is highly recommended to avoid unwanted artifacts. You can use the negative prompt suggested in the original RealVis XL model card.
  • Hires Fix: Consider using a hires fix (upscaling and denoising) for even better results, as suggested in the RealVis XL model card.
  • Experiment with LoRA Scale: Adjust the lora_scale parameter to control the influence of the Life FX LoRA on the final image.

Limitations:

While this model is very capable, it may still exhibit some of the limitations inherent in both the base model and the LoRA, such as difficulties with complex compositions or unusual prompts.

Acknowledgments: