qwen-image-realism-lora

LORA Qwen-Image example

Lora for Qwen-Image

Realism word required

📝 Updates

August 13, 2025 - Realism v1.1 Release

• Enhanced facial detail quality: Improved fine-grained facial features and skin texture rendering
• Better color reproduction: More accurate and vibrant color representation across all scenes
• Improved lighting and shadows for landscapes: Enhanced natural lighting effects and shadow detail in outdoor scenes
• Increased diversity across different ethnicities: Better representation and quality for various racial and ethnic features

🧪 Usage

---

🔧 Initialization

from diffusers import DiffusionPipeline
import torch

model_name = "Qwen/Qwen-Image"

# Load the pipeline
if torch.cuda.is_available():
    torch_dtype = torch.bfloat16
    device = "cuda"
else:
    torch_dtype = torch.float32
    device = "cpu"

pipe = DiffusionPipeline.from_pretrained(model_name, torch_dtype=torch_dtype)
pipe = pipe.to(device)

🔌 Load LoRA Weights

# Load LoRA weights
pipe.load_lora_weights('flymy_realism.safetensors', adapter_name="lora")

🎨 Generate Image with lora trained on person

prompt = '''Super Realism portrait of a teenager woman of African descent, serene calmness, arms crossed, illuminated by dramatic studio lighting, sunlit park in the background, adorned with delicate jewelry, three-quarter view, sun-kissed skin with natural imperfections, loose shoulder-length curls, slightly squinting eyes, environmental street portrait with text "FLYMY AI" on t-shirt.'''
negative_prompt =  " "
image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=1024,
    height=1024,
    num_inference_steps=50,
    true_cfg_scale=5,
    generator=torch.Generator(device="cuda").manual_seed(346346)
)

# Display the image (in Jupyter or save to file)
image.show()
# or
image.save("output.png")

🖼️ Sample Output

🎛️ Using with ComfyUI

We provide a ready-to-use ComfyUI workflow that works with our trained LoRA models. Follow these steps to set up and use the workflow:

Setup Instructions

1. Download the latest ComfyUI:

   • Visit the ComfyUI GitHub repository
   • Clone or download the latest version

2. Install ComfyUI:

   • Follow the installation instructions from the ComfyUI repository
   • Make sure all dependencies are properly installed

3. Download Qwen-Image model weights:

   • Go to Qwen-Image ComfyUI weights
   • Download all the model files

4. Place Qwen-Image weights in ComfyUI:

   • Copy the downloaded Qwen-Image model files to the appropriate folders in ComfyUI/models/
   • Follow the folder structure as specified in the model repository

5. Download our pre-trained LoRA weights:

   • Visit flymy-ai/qwen-image-lora
   • Download the LoRA .safetensors files

6. Place LoRA weights in ComfyUI:

   • Copy the LoRA file flymy-ai/qwen-image-lora/pytorch_lora_weights.safetensors to ComfyUI/models/loras/

7. Load the workflow:

   • Open ComfyUI in your browser
   • Load the workflow file qwen_image_lora_example.json located in this repository
   • The workflow is pre-configured to work with our LoRA models

Workflow Features

• ✅ Pre-configured for Qwen-Image + LoRA inference
• ✅ Optimized settings for best quality output
• ✅ Easy prompt and parameter adjustment
• ✅ Compatible with all our trained LoRA models

The ComfyUI workflow provides a user-friendly interface for generating images with our trained LoRA models without needing to write Python code.

🖼️ Workflow Screenshot

📊 Side-by-Side Comparisons

Below are comparison images showing the improvements of our LoRA model across different scenarios:

Mountain Landscape

Prompt: "realism, mountain landscape at golden hour, low sun grazing alpine meadows, crisp ridgelines, thin mist in valley, high dynamic range sky, 24mm f/8, ISO 100, tripod, ultra-sharp foreground grass, micro-contrast on rocks"

Rainy Night City Scene

Prompt: "realism, rainy night city scene with neon reflections, person holding a transparent umbrella, water droplets sharp on umbrella surface, shallow DOF, 55mm f/1.8, ISO 1600, blue and magenta neon, storefront sign reads "OPEN 24/7""

Athletic Motion

Prompt: "realism, athlete mid-sprint on track, strong sunlight, backlit dust particles, frozen motion at 1/2000s, 200mm f/2.8, ISO 400, muscle definition and sweat droplets detailed, stadium banner says "FINALS""

Corporate Portrait

Prompt: "realism, corporate headshot of a CTO in glass-walled office, city skyline bokeh, balanced key/fill lighting, 85mm f/2, ISO 100, crisp lapel and hair detail, subtle reflection on glasses"

🤝 Support

If you have questions or suggestions, join our community:

• 🌐 FlyMy.AI
• 💬 Discord Community
• 🐦 Follow us on X
• 💼 Connect on LinkedIn
• 📧 Support

⭐ Don't forget to star the repository if you like it!

---

license: apache-2.0